diff --git "a/nvidia_parakeet-v2_333MB/AudioEncoder.mlmodelc/model.mil" "b/nvidia_parakeet-v2_333MB/AudioEncoder.mlmodelc/model.mil" deleted file mode 100644--- "a/nvidia_parakeet-v2_333MB/AudioEncoder.mlmodelc/model.mil" +++ /dev/null @@ -1,6617 +0,0 @@ -program(1.3) -[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}})] -{ - func main(tensor melspectrogram_features) { - string input_1_pad_type_0 = const()[name = string("input_1_pad_type_0"), val = string("custom")]; - tensor input_1_pad_0 = const()[name = string("input_1_pad_0"), val = tensor([1, 1, 1, 1])]; - tensor input_1_strides_0 = const()[name = string("input_1_strides_0"), val = tensor([2, 2])]; - tensor input_1_dilations_0 = const()[name = string("input_1_dilations_0"), val = tensor([1, 1])]; - int32 input_1_groups_0 = const()[name = string("input_1_groups_0"), val = int32(1)]; - tensor pre_encode_conv_0_weight_to_fp16 = const()[name = string("pre_encode_conv_0_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; - tensor pre_encode_conv_0_bias_to_fp16 = const()[name = string("pre_encode_conv_0_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))]; - tensor input_1_cast_fp16 = conv(bias = pre_encode_conv_0_bias_to_fp16, dilations = input_1_dilations_0, groups = input_1_groups_0, pad = input_1_pad_0, pad_type = input_1_pad_type_0, strides = input_1_strides_0, weight = pre_encode_conv_0_weight_to_fp16, x = melspectrogram_features)[name = string("input_1_cast_fp16")]; - tensor input_3_cast_fp16 = relu(x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; - string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("custom")]; - tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([1, 1, 1, 1])]; - tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([2, 2])]; - int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(256)]; - tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; - tensor pre_encode_conv_2_weight_to_fp16 = const()[name = string("pre_encode_conv_2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))]; - tensor pre_encode_conv_2_bias_to_fp16 = const()[name = string("pre_encode_conv_2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))]; - tensor input_5_cast_fp16 = conv(bias = pre_encode_conv_2_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = pre_encode_conv_2_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; - string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; - tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; - tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; - int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; - tensor pre_encode_conv_3_weight_to_fp16 = const()[name = string("pre_encode_conv_3_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))]; - tensor pre_encode_conv_3_bias_to_fp16 = const()[name = string("pre_encode_conv_3_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))]; - tensor input_7_cast_fp16 = conv(bias = pre_encode_conv_3_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = pre_encode_conv_3_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; - tensor input_9_cast_fp16 = relu(x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; - string input_11_pad_type_0 = const()[name = string("input_11_pad_type_0"), val = string("custom")]; - tensor input_11_pad_0 = const()[name = string("input_11_pad_0"), val = tensor([1, 1, 1, 1])]; - tensor input_11_strides_0 = const()[name = string("input_11_strides_0"), val = tensor([2, 2])]; - int32 input_11_groups_0 = const()[name = string("input_11_groups_0"), val = int32(256)]; - tensor input_11_dilations_0 = const()[name = string("input_11_dilations_0"), val = tensor([1, 1])]; - tensor pre_encode_conv_5_weight_to_fp16 = const()[name = string("pre_encode_conv_5_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))]; - tensor pre_encode_conv_5_bias_to_fp16 = const()[name = string("pre_encode_conv_5_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))]; - tensor input_11_cast_fp16 = conv(bias = pre_encode_conv_5_bias_to_fp16, dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = pre_encode_conv_5_weight_to_fp16, x = input_9_cast_fp16)[name = string("input_11_cast_fp16")]; - string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; - tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; - tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; - int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; - tensor pre_encode_conv_6_weight_to_fp16 = const()[name = string("pre_encode_conv_6_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))]; - tensor pre_encode_conv_6_bias_to_fp16 = const()[name = string("pre_encode_conv_6_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))]; - tensor input_13_cast_fp16 = conv(bias = pre_encode_conv_6_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = pre_encode_conv_6_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; - tensor x_1_cast_fp16 = relu(x = input_13_cast_fp16)[name = string("x_1_cast_fp16")]; - tensor var_112_perm_0 = const()[name = string("op_112_perm_0"), val = tensor([0, 1, 3, 2])]; - tensor var_115 = const()[name = string("op_115"), val = tensor([1, 4096, 1, 188])]; - tensor var_112_cast_fp16 = transpose(perm = var_112_perm_0, x = x_1_cast_fp16)[name = string("transpose_0")]; - tensor input_15_cast_fp16 = reshape(shape = var_115, x = var_112_cast_fp16)[name = string("input_15_cast_fp16")]; - string var_125_pad_type_0 = const()[name = string("op_125_pad_type_0"), val = string("valid")]; - tensor var_125_strides_0 = const()[name = string("op_125_strides_0"), val = tensor([1, 1])]; - tensor var_125_pad_0 = const()[name = string("op_125_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_125_dilations_0 = const()[name = string("op_125_dilations_0"), val = tensor([1, 1])]; - int32 var_125_groups_0 = const()[name = string("op_125_groups_0"), val = int32(1)]; - tensor pre_encode_out_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1852160))))[name = string("pre_encode_out_inlier_module_weight_to_fp16_palettized")]; - tensor pre_encode_out_inlier_module_bias_to_fp16 = const()[name = string("pre_encode_out_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1853248)))]; - tensor var_125_cast_fp16 = conv(bias = pre_encode_out_inlier_module_bias_to_fp16, dilations = var_125_dilations_0, groups = var_125_groups_0, pad = var_125_pad_0, pad_type = var_125_pad_type_0, strides = var_125_strides_0, weight = pre_encode_out_inlier_module_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = string("op_125_cast_fp16")]; - string var_131_pad_type_0 = const()[name = string("op_131_pad_type_0"), val = string("valid")]; - tensor var_131_strides_0 = const()[name = string("op_131_strides_0"), val = tensor([1, 1])]; - tensor var_131_pad_0 = const()[name = string("op_131_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_131_dilations_0 = const()[name = string("op_131_dilations_0"), val = tensor([1, 1])]; - int32 var_131_groups_0 = const()[name = string("op_131_groups_0"), val = int32(1)]; - tensor pre_encode_out_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1915840))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1855360))))[name = string("pre_encode_out_outlier_module_weight_to_fp16_sparsified")]; - tensor var_131_cast_fp16 = conv(dilations = var_131_dilations_0, groups = var_131_groups_0, pad = var_131_pad_0, pad_type = var_131_pad_type_0, strides = var_131_strides_0, weight = pre_encode_out_outlier_module_weight_to_fp16_sparsified, x = input_15_cast_fp16)[name = string("op_131_cast_fp16")]; - tensor inputs_1_cast_fp16 = add(x = var_125_cast_fp16, y = var_131_cast_fp16)[name = string("inputs_1_cast_fp16")]; - int32 var_137 = const()[name = string("op_137"), val = int32(3)]; - tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; - fp16 var_168_to_fp16 = const()[name = string("op_168_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_168_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; - tensor input_17_mean_0_to_fp16 = const()[name = string("input_17_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2440192)))]; - tensor input_17_variance_0_to_fp16 = const()[name = string("input_17_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2442304)))]; - tensor input_17_gamma_0_to_fp16 = const()[name = string("input_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2444416)))]; - tensor input_17_beta_0_to_fp16 = const()[name = string("input_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2446528)))]; - fp16 input_17_epsilon_0_to_fp16 = const()[name = string("input_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_17_cast_fp16 = batch_norm(beta = input_17_beta_0_to_fp16, epsilon = input_17_epsilon_0_to_fp16, gamma = input_17_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("input_17_cast_fp16")]; - string var_188_pad_type_0 = const()[name = string("op_188_pad_type_0"), val = string("valid")]; - tensor var_188_strides_0 = const()[name = string("op_188_strides_0"), val = tensor([1, 1])]; - tensor var_188_pad_0 = const()[name = string("op_188_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_188_dilations_0 = const()[name = string("op_188_dilations_0"), val = tensor([1, 1])]; - int32 var_188_groups_0 = const()[name = string("op_188_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2448640))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4021568))))[name = string("layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = string("layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4025728)))]; - tensor var_188_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_188_dilations_0, groups = var_188_groups_0, pad = var_188_pad_0, pad_type = var_188_pad_type_0, strides = var_188_strides_0, weight = layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = string("op_188_cast_fp16")]; - string var_194_pad_type_0 = const()[name = string("op_194_pad_type_0"), val = string("valid")]; - tensor var_194_strides_0 = const()[name = string("op_194_strides_0"), val = tensor([1, 1])]; - tensor var_194_pad_0 = const()[name = string("op_194_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_194_dilations_0 = const()[name = string("op_194_dilations_0"), val = tensor([1, 1])]; - int32 var_194_groups_0 = const()[name = string("op_194_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4102848))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4033984))))[name = string("layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_194_cast_fp16 = conv(dilations = var_194_dilations_0, groups = var_194_groups_0, pad = var_194_pad_0, pad_type = var_194_pad_type_0, strides = var_194_strides_0, weight = layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_17_cast_fp16)[name = string("op_194_cast_fp16")]; - tensor input_19_cast_fp16 = add(x = var_188_cast_fp16, y = var_194_cast_fp16)[name = string("input_19_cast_fp16")]; - tensor input_21_cast_fp16 = silu(x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; - string var_205_pad_type_0 = const()[name = string("op_205_pad_type_0"), val = string("valid")]; - tensor var_205_strides_0 = const()[name = string("op_205_strides_0"), val = tensor([1, 1])]; - tensor var_205_pad_0 = const()[name = string("op_205_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_205_dilations_0 = const()[name = string("op_205_dilations_0"), val = tensor([1, 1])]; - int32 var_205_groups_0 = const()[name = string("op_205_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4627200))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6200128))))[name = string("layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_205_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_205_dilations_0, groups = var_205_groups_0, pad = var_205_pad_0, pad_type = var_205_pad_type_0, strides = var_205_strides_0, weight = layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = string("op_205_cast_fp16")]; - string var_211_pad_type_0 = const()[name = string("op_211_pad_type_0"), val = string("valid")]; - tensor var_211_strides_0 = const()[name = string("op_211_strides_0"), val = tensor([1, 1])]; - tensor var_211_pad_0 = const()[name = string("op_211_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_211_dilations_0 = const()[name = string("op_211_dilations_0"), val = tensor([1, 1])]; - int32 var_211_groups_0 = const()[name = string("op_211_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6268480))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6201216))))[name = string("layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_211_cast_fp16 = conv(dilations = var_211_dilations_0, groups = var_211_groups_0, pad = var_211_pad_0, pad_type = var_211_pad_type_0, strides = var_211_strides_0, weight = layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_21_cast_fp16)[name = string("op_211_cast_fp16")]; - tensor x_3_cast_fp16 = add(x = var_205_cast_fp16, y = var_211_cast_fp16)[name = string("x_3_cast_fp16")]; - fp16 var_213_to_fp16 = const()[name = string("op_213_to_fp16"), val = fp16(0x1p-1)]; - tensor var_214_cast_fp16 = mul(x = x_3_cast_fp16, y = var_213_to_fp16)[name = string("op_214_cast_fp16")]; - tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = var_214_cast_fp16)[name = string("inputs_3_cast_fp16")]; - tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; - fp16 var_224_to_fp16 = const()[name = string("op_224_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_224_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; - tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6792832)))]; - tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6794944)))]; - fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_1_cast_fp16")]; - string var_249_pad_type_0 = const()[name = string("op_249_pad_type_0"), val = string("valid")]; - tensor var_249_strides_0 = const()[name = string("op_249_strides_0"), val = tensor([1, 1])]; - tensor var_249_pad_0 = const()[name = string("op_249_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_249_dilations_0 = const()[name = string("op_249_dilations_0"), val = tensor([1, 1])]; - int32 var_249_groups_0 = const()[name = string("op_249_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6797056))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7190336))))[name = string("layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_249_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_249_dilations_0, groups = var_249_groups_0, pad = var_249_pad_0, pad_type = var_249_pad_type_0, strides = var_249_strides_0, weight = layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = string("op_249_cast_fp16")]; - string var_255_pad_type_0 = const()[name = string("op_255_pad_type_0"), val = string("valid")]; - tensor var_255_strides_0 = const()[name = string("op_255_strides_0"), val = tensor([1, 1])]; - tensor var_255_pad_0 = const()[name = string("op_255_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_255_dilations_0 = const()[name = string("op_255_dilations_0"), val = tensor([1, 1])]; - int32 var_255_groups_0 = const()[name = string("op_255_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7203968))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7191424))))[name = string("layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_255_cast_fp16 = conv(dilations = var_255_dilations_0, groups = var_255_groups_0, pad = var_255_pad_0, pad_type = var_255_pad_type_0, strides = var_255_strides_0, weight = layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = string("op_255_cast_fp16")]; - tensor query_1_cast_fp16 = add(x = var_249_cast_fp16, y = var_255_cast_fp16)[name = string("query_1_cast_fp16")]; - string var_264_pad_type_0 = const()[name = string("op_264_pad_type_0"), val = string("valid")]; - tensor var_264_strides_0 = const()[name = string("op_264_strides_0"), val = tensor([1, 1])]; - tensor var_264_pad_0 = const()[name = string("op_264_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_264_dilations_0 = const()[name = string("op_264_dilations_0"), val = tensor([1, 1])]; - int32 var_264_groups_0 = const()[name = string("op_264_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7335104))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7728384))))[name = string("layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_264_cast_fp16 = conv(dilations = var_264_dilations_0, groups = var_264_groups_0, pad = var_264_pad_0, pad_type = var_264_pad_type_0, strides = var_264_strides_0, weight = layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = string("op_264_cast_fp16")]; - string var_270_pad_type_0 = const()[name = string("op_270_pad_type_0"), val = string("valid")]; - tensor var_270_strides_0 = const()[name = string("op_270_strides_0"), val = tensor([1, 1])]; - tensor var_270_pad_0 = const()[name = string("op_270_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_270_dilations_0 = const()[name = string("op_270_dilations_0"), val = tensor([1, 1])]; - int32 var_270_groups_0 = const()[name = string("op_270_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7752768))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7729472))))[name = string("layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_270_cast_fp16 = conv(dilations = var_270_dilations_0, groups = var_270_groups_0, pad = var_270_pad_0, pad_type = var_270_pad_type_0, strides = var_270_strides_0, weight = layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = string("op_270_cast_fp16")]; - tensor key_1_cast_fp16 = add(x = var_264_cast_fp16, y = var_270_cast_fp16)[name = string("key_1_cast_fp16")]; - string var_280_pad_type_0 = const()[name = string("op_280_pad_type_0"), val = string("valid")]; - tensor var_280_strides_0 = const()[name = string("op_280_strides_0"), val = tensor([1, 1])]; - tensor var_280_pad_0 = const()[name = string("op_280_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_280_dilations_0 = const()[name = string("op_280_dilations_0"), val = tensor([1, 1])]; - int32 var_280_groups_0 = const()[name = string("op_280_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7883904))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8277184))))[name = string("layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_280_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_280_dilations_0, groups = var_280_groups_0, pad = var_280_pad_0, pad_type = var_280_pad_type_0, strides = var_280_strides_0, weight = layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = string("op_280_cast_fp16")]; - string var_286_pad_type_0 = const()[name = string("op_286_pad_type_0"), val = string("valid")]; - tensor var_286_strides_0 = const()[name = string("op_286_strides_0"), val = tensor([1, 1])]; - tensor var_286_pad_0 = const()[name = string("op_286_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_286_dilations_0 = const()[name = string("op_286_dilations_0"), val = tensor([1, 1])]; - int32 var_286_groups_0 = const()[name = string("op_286_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8292480))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8278272))))[name = string("layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_286_cast_fp16 = conv(dilations = var_286_dilations_0, groups = var_286_groups_0, pad = var_286_pad_0, pad_type = var_286_pad_type_0, strides = var_286_strides_0, weight = layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = string("op_286_cast_fp16")]; - tensor value_1_cast_fp16 = add(x = var_280_cast_fp16, y = var_286_cast_fp16)[name = string("value_1_cast_fp16")]; - tensor var_289_to_fp16 = const()[name = string("op_289_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8423616)))]; - tensor query_3_cast_fp16 = add(x = query_1_cast_fp16, y = var_289_to_fp16)[name = string("query_3_cast_fp16")]; - tensor var_292_to_fp16 = const()[name = string("op_292_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8425728)))]; - tensor q_with_bias_v_1_cast_fp16 = add(x = query_1_cast_fp16, y = var_292_to_fp16)[name = string("q_with_bias_v_1_cast_fp16")]; - string var_302_pad_type_0 = const()[name = string("op_302_pad_type_0"), val = string("valid")]; - tensor var_302_strides_0 = const()[name = string("op_302_strides_0"), val = tensor([1, 1])]; - tensor var_302_pad_0 = const()[name = string("op_302_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_302_dilations_0 = const()[name = string("op_302_dilations_0"), val = tensor([1, 1])]; - int32 var_302_groups_0 = const()[name = string("op_302_groups_0"), val = int32(1)]; - tensor pos_enc_to_fp16 = const()[name = string("pos_enc_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8427840)))]; - tensor layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9195904))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9589184))))[name = string("layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_302_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_302_dilations_0, groups = var_302_groups_0, pad = var_302_pad_0, pad_type = var_302_pad_type_0, strides = var_302_strides_0, weight = layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_302_cast_fp16")]; - string var_308_pad_type_0 = const()[name = string("op_308_pad_type_0"), val = string("valid")]; - tensor var_308_strides_0 = const()[name = string("op_308_strides_0"), val = tensor([1, 1])]; - tensor var_308_pad_0 = const()[name = string("op_308_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_308_dilations_0 = const()[name = string("op_308_dilations_0"), val = tensor([1, 1])]; - int32 var_308_groups_0 = const()[name = string("op_308_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9629760))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9590272))))[name = string("layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_308_cast_fp16 = conv(dilations = var_308_dilations_0, groups = var_308_groups_0, pad = var_308_pad_0, pad_type = var_308_pad_type_0, strides = var_308_strides_0, weight = layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_308_cast_fp16")]; - tensor p_1_cast_fp16 = add(x = var_302_cast_fp16, y = var_308_cast_fp16)[name = string("p_1_cast_fp16")]; - tensor var_312 = const()[name = string("op_312"), val = tensor([1, 8, 128, 188])]; - tensor var_313_cast_fp16 = reshape(shape = var_312, x = q_with_bias_v_1_cast_fp16)[name = string("op_313_cast_fp16")]; - tensor var_314 = const()[name = string("op_314"), val = tensor([1, 8, 128, -1])]; - tensor var_315_cast_fp16 = reshape(shape = var_314, x = p_1_cast_fp16)[name = string("op_315_cast_fp16")]; - bool matrix_bd_1_transpose_x_0 = const()[name = string("matrix_bd_1_transpose_x_0"), val = bool(true)]; - bool matrix_bd_1_transpose_y_0 = const()[name = string("matrix_bd_1_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_1_cast_fp16 = matmul(transpose_x = matrix_bd_1_transpose_x_0, transpose_y = matrix_bd_1_transpose_y_0, x = var_313_cast_fp16, y = var_315_cast_fp16)[name = string("matrix_bd_1_cast_fp16")]; - tensor matrix_bd_3_pad_0 = const()[name = string("matrix_bd_3_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_3_mode_0 = const()[name = string("matrix_bd_3_mode_0"), val = string("constant")]; - fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_3_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = matrix_bd_3_mode_0, pad = matrix_bd_3_pad_0, x = matrix_bd_1_cast_fp16)[name = string("matrix_bd_3_cast_fp16")]; - tensor var_324 = const()[name = string("op_324"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_5_cast_fp16 = reshape(shape = var_324, x = matrix_bd_3_cast_fp16)[name = string("matrix_bd_5_cast_fp16")]; - tensor var_328_begin_0 = const()[name = string("op_328_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_328_end_0 = const()[name = string("op_328_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_328_end_mask_0 = const()[name = string("op_328_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_328_cast_fp16 = slice_by_index(begin = var_328_begin_0, end = var_328_end_0, end_mask = var_328_end_mask_0, x = matrix_bd_5_cast_fp16)[name = string("op_328_cast_fp16")]; - tensor var_329 = const()[name = string("op_329"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_7_cast_fp16 = reshape(shape = var_329, x = var_328_cast_fp16)[name = string("matrix_bd_7_cast_fp16")]; - tensor var_334_begin_0 = const()[name = string("op_334_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_334_end_0 = const()[name = string("op_334_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_334_end_mask_0 = const()[name = string("op_334_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_334_cast_fp16 = slice_by_index(begin = var_334_begin_0, end = var_334_end_0, end_mask = var_334_end_mask_0, x = matrix_bd_7_cast_fp16)[name = string("op_334_cast_fp16")]; - fp16 var_335_to_fp16 = const()[name = string("op_335_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_1_cast_fp16 = mul(x = var_334_cast_fp16, y = var_335_to_fp16)[name = string("qk_mask_1_cast_fp16")]; - tensor var_339 = const()[name = string("op_339"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_1_cast_fp16 = reshape(shape = var_339, x = query_3_cast_fp16)[name = string("mh_q_1_cast_fp16")]; - fp16 var_341_to_fp16 = const()[name = string("op_341_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_342_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_341_to_fp16)[name = string("op_342_cast_fp16")]; - tensor var_345 = const()[name = string("op_345"), val = tensor([1, 8, 128, 188])]; - tensor var_346_cast_fp16 = reshape(shape = var_345, x = key_1_cast_fp16)[name = string("op_346_cast_fp16")]; - bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; - bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; - tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_342_cast_fp16, y = var_346_cast_fp16)[name = string("mh_w_1_cast_fp16")]; - tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = qk_mask_1_cast_fp16)[name = string("mh_w_3_cast_fp16")]; - tensor var_350_cast_fp16 = softmax(axis = var_137, x = mh_w_3_cast_fp16)[name = string("op_350_cast_fp16")]; - tensor var_351 = const()[name = string("op_351"), val = tensor([1, 8, 128, 188])]; - tensor var_352_cast_fp16 = reshape(shape = var_351, x = value_1_cast_fp16)[name = string("op_352_cast_fp16")]; - bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; - bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; - tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_352_cast_fp16, y = var_350_cast_fp16)[name = string("attn_1_cast_fp16")]; - tensor var_355 = const()[name = string("op_355"), val = tensor([1, 1024, 1, 188])]; - tensor input_23_cast_fp16 = reshape(shape = var_355, x = attn_1_cast_fp16)[name = string("input_23_cast_fp16")]; - string var_365_pad_type_0 = const()[name = string("op_365_pad_type_0"), val = string("valid")]; - tensor var_365_strides_0 = const()[name = string("op_365_strides_0"), val = tensor([1, 1])]; - tensor var_365_pad_0 = const()[name = string("op_365_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_365_dilations_0 = const()[name = string("op_365_dilations_0"), val = tensor([1, 1])]; - int32 var_365_groups_0 = const()[name = string("op_365_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9760896))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10154176))))[name = string("layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_365_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_365_dilations_0, groups = var_365_groups_0, pad = var_365_pad_0, pad_type = var_365_pad_type_0, strides = var_365_strides_0, weight = layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_23_cast_fp16)[name = string("op_365_cast_fp16")]; - string var_371_pad_type_0 = const()[name = string("op_371_pad_type_0"), val = string("valid")]; - tensor var_371_strides_0 = const()[name = string("op_371_strides_0"), val = tensor([1, 1])]; - tensor var_371_pad_0 = const()[name = string("op_371_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_371_dilations_0 = const()[name = string("op_371_dilations_0"), val = tensor([1, 1])]; - int32 var_371_groups_0 = const()[name = string("op_371_groups_0"), val = int32(1)]; - tensor layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10168000))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10155264))))[name = string("layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_371_cast_fp16 = conv(dilations = var_371_dilations_0, groups = var_371_groups_0, pad = var_371_pad_0, pad_type = var_371_pad_type_0, strides = var_371_strides_0, weight = layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_23_cast_fp16)[name = string("op_371_cast_fp16")]; - tensor obj_5_cast_fp16 = add(x = var_365_cast_fp16, y = var_371_cast_fp16)[name = string("obj_5_cast_fp16")]; - tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; - tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; - fp16 var_382_to_fp16 = const()[name = string("op_382_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_382_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; - tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10299136)))]; - tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10301248)))]; - fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_25_cast_fp16")]; - string var_403_pad_type_0 = const()[name = string("op_403_pad_type_0"), val = string("valid")]; - tensor var_403_strides_0 = const()[name = string("op_403_strides_0"), val = tensor([1, 1])]; - tensor var_403_pad_0 = const()[name = string("op_403_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_403_dilations_0 = const()[name = string("op_403_dilations_0"), val = tensor([1, 1])]; - int32 var_403_groups_0 = const()[name = string("op_403_groups_0"), val = int32(1)]; - tensor layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10303360))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11089856))))[name = string("layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_403_cast_fp16 = conv(dilations = var_403_dilations_0, groups = var_403_groups_0, pad = var_403_pad_0, pad_type = var_403_pad_type_0, strides = var_403_strides_0, weight = layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = string("op_403_cast_fp16")]; - string var_409_pad_type_0 = const()[name = string("op_409_pad_type_0"), val = string("valid")]; - tensor var_409_strides_0 = const()[name = string("op_409_strides_0"), val = tensor([1, 1])]; - tensor var_409_pad_0 = const()[name = string("op_409_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_409_dilations_0 = const()[name = string("op_409_dilations_0"), val = tensor([1, 1])]; - int32 var_409_groups_0 = const()[name = string("op_409_groups_0"), val = int32(1)]; - tensor layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11116352))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11091968))))[name = string("layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_409_cast_fp16 = conv(dilations = var_409_dilations_0, groups = var_409_groups_0, pad = var_409_pad_0, pad_type = var_409_pad_type_0, strides = var_409_strides_0, weight = layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_25_cast_fp16)[name = string("op_409_cast_fp16")]; - tensor input_27_cast_fp16 = add(x = var_403_cast_fp16, y = var_409_cast_fp16)[name = string("input_27_cast_fp16")]; - int32 input_29_split_num_splits_0 = const()[name = string("input_29_split_num_splits_0"), val = int32(2)]; - int32 input_29_split_axis_0 = const()[name = string("input_29_split_axis_0"), val = int32(1)]; - tensor input_29_split_cast_fp16_0, tensor input_29_split_cast_fp16_1 = split(axis = input_29_split_axis_0, num_splits = input_29_split_num_splits_0, x = input_27_cast_fp16)[name = string("input_29_split_cast_fp16")]; - tensor input_29_split_1_sigmoid_cast_fp16 = sigmoid(x = input_29_split_cast_fp16_1)[name = string("input_29_split_1_sigmoid_cast_fp16")]; - tensor input_29_cast_fp16 = mul(x = input_29_split_cast_fp16_0, y = input_29_split_1_sigmoid_cast_fp16)[name = string("input_29_cast_fp16")]; - string input_31_pad_type_0 = const()[name = string("input_31_pad_type_0"), val = string("custom")]; - tensor input_31_pad_0 = const()[name = string("input_31_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_31_groups_0 = const()[name = string("input_31_groups_0"), val = int32(1024)]; - tensor input_31_strides_0 = const()[name = string("input_31_strides_0"), val = tensor([1, 1])]; - tensor input_31_dilations_0 = const()[name = string("input_31_dilations_0"), val = tensor([1, 1])]; - tensor const_268_to_fp16 = const()[name = string("const_268_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11378560)))]; - tensor const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11397056)))]; - tensor input_33_cast_fp16 = conv(bias = const_269_to_fp16, dilations = input_31_dilations_0, groups = input_31_groups_0, pad = input_31_pad_0, pad_type = input_31_pad_type_0, strides = input_31_strides_0, weight = const_268_to_fp16, x = input_29_cast_fp16)[name = string("input_33_cast_fp16")]; - tensor input_35_cast_fp16 = silu(x = input_33_cast_fp16)[name = string("input_35_cast_fp16")]; - string var_431_pad_type_0 = const()[name = string("op_431_pad_type_0"), val = string("valid")]; - tensor var_431_strides_0 = const()[name = string("op_431_strides_0"), val = tensor([1, 1])]; - tensor var_431_pad_0 = const()[name = string("op_431_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_431_dilations_0 = const()[name = string("op_431_dilations_0"), val = tensor([1, 1])]; - int32 var_431_groups_0 = const()[name = string("op_431_groups_0"), val = int32(1)]; - tensor layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11399168))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11792448))))[name = string("layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_431_cast_fp16 = conv(dilations = var_431_dilations_0, groups = var_431_groups_0, pad = var_431_pad_0, pad_type = var_431_pad_type_0, strides = var_431_strides_0, weight = layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = string("op_431_cast_fp16")]; - string var_437_pad_type_0 = const()[name = string("op_437_pad_type_0"), val = string("valid")]; - tensor var_437_strides_0 = const()[name = string("op_437_strides_0"), val = tensor([1, 1])]; - tensor var_437_pad_0 = const()[name = string("op_437_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_437_dilations_0 = const()[name = string("op_437_dilations_0"), val = tensor([1, 1])]; - int32 var_437_groups_0 = const()[name = string("op_437_groups_0"), val = int32(1)]; - tensor layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11806592))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11793536))))[name = string("layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_437_cast_fp16 = conv(dilations = var_437_dilations_0, groups = var_437_groups_0, pad = var_437_pad_0, pad_type = var_437_pad_type_0, strides = var_437_strides_0, weight = layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_35_cast_fp16)[name = string("op_437_cast_fp16")]; - tensor x_5_cast_fp16 = add(x = var_431_cast_fp16, y = var_437_cast_fp16)[name = string("x_5_cast_fp16")]; - tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = x_5_cast_fp16)[name = string("inputs_7_cast_fp16")]; - tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; - fp16 var_448_to_fp16 = const()[name = string("op_448_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_448_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; - tensor input_37_gamma_0_to_fp16 = const()[name = string("input_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11937728)))]; - tensor input_37_beta_0_to_fp16 = const()[name = string("input_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11939840)))]; - fp16 input_37_epsilon_0_to_fp16 = const()[name = string("input_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_37_cast_fp16 = batch_norm(beta = input_37_beta_0_to_fp16, epsilon = input_37_epsilon_0_to_fp16, gamma = input_37_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_37_cast_fp16")]; - string var_468_pad_type_0 = const()[name = string("op_468_pad_type_0"), val = string("valid")]; - tensor var_468_strides_0 = const()[name = string("op_468_strides_0"), val = tensor([1, 1])]; - tensor var_468_pad_0 = const()[name = string("op_468_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_468_dilations_0 = const()[name = string("op_468_dilations_0"), val = tensor([1, 1])]; - int32 var_468_groups_0 = const()[name = string("op_468_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11941952))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13514880))))[name = string("layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_468_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_468_dilations_0, groups = var_468_groups_0, pad = var_468_pad_0, pad_type = var_468_pad_type_0, strides = var_468_strides_0, weight = layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = string("op_468_cast_fp16")]; - string var_474_pad_type_0 = const()[name = string("op_474_pad_type_0"), val = string("valid")]; - tensor var_474_strides_0 = const()[name = string("op_474_strides_0"), val = tensor([1, 1])]; - tensor var_474_pad_0 = const()[name = string("op_474_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_474_dilations_0 = const()[name = string("op_474_dilations_0"), val = tensor([1, 1])]; - int32 var_474_groups_0 = const()[name = string("op_474_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13565760))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13519040))))[name = string("layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_474_cast_fp16 = conv(dilations = var_474_dilations_0, groups = var_474_groups_0, pad = var_474_pad_0, pad_type = var_474_pad_type_0, strides = var_474_strides_0, weight = layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_37_cast_fp16)[name = string("op_474_cast_fp16")]; - tensor input_39_cast_fp16 = add(x = var_468_cast_fp16, y = var_474_cast_fp16)[name = string("input_39_cast_fp16")]; - tensor input_41_cast_fp16 = silu(x = input_39_cast_fp16)[name = string("input_41_cast_fp16")]; - string var_485_pad_type_0 = const()[name = string("op_485_pad_type_0"), val = string("valid")]; - tensor var_485_strides_0 = const()[name = string("op_485_strides_0"), val = tensor([1, 1])]; - tensor var_485_pad_0 = const()[name = string("op_485_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_485_dilations_0 = const()[name = string("op_485_dilations_0"), val = tensor([1, 1])]; - int32 var_485_groups_0 = const()[name = string("op_485_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14090112))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15663040))))[name = string("layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_485_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_485_dilations_0, groups = var_485_groups_0, pad = var_485_pad_0, pad_type = var_485_pad_type_0, strides = var_485_strides_0, weight = layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = string("op_485_cast_fp16")]; - string var_491_pad_type_0 = const()[name = string("op_491_pad_type_0"), val = string("valid")]; - tensor var_491_strides_0 = const()[name = string("op_491_strides_0"), val = tensor([1, 1])]; - tensor var_491_pad_0 = const()[name = string("op_491_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_491_dilations_0 = const()[name = string("op_491_dilations_0"), val = tensor([1, 1])]; - int32 var_491_groups_0 = const()[name = string("op_491_groups_0"), val = int32(1)]; - tensor layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15714176))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15664128))))[name = string("layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_491_cast_fp16 = conv(dilations = var_491_dilations_0, groups = var_491_groups_0, pad = var_491_pad_0, pad_type = var_491_pad_type_0, strides = var_491_strides_0, weight = layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_41_cast_fp16)[name = string("op_491_cast_fp16")]; - tensor x_7_cast_fp16 = add(x = var_485_cast_fp16, y = var_491_cast_fp16)[name = string("x_7_cast_fp16")]; - fp16 var_493_to_fp16 = const()[name = string("op_493_to_fp16"), val = fp16(0x1p-1)]; - tensor var_494_cast_fp16 = mul(x = x_7_cast_fp16, y = var_493_to_fp16)[name = string("op_494_cast_fp16")]; - tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = var_494_cast_fp16)[name = string("inputs_9_cast_fp16")]; - tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; - fp16 var_504_to_fp16 = const()[name = string("op_504_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_504_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; - tensor inputs_11_gamma_0_to_fp16 = const()[name = string("inputs_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16238528)))]; - tensor inputs_11_beta_0_to_fp16 = const()[name = string("inputs_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16240640)))]; - fp16 inputs_11_epsilon_0_to_fp16 = const()[name = string("inputs_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_11_cast_fp16 = batch_norm(beta = inputs_11_beta_0_to_fp16, epsilon = inputs_11_epsilon_0_to_fp16, gamma = inputs_11_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("inputs_11_cast_fp16")]; - int32 var_518 = const()[name = string("op_518"), val = int32(3)]; - tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; - fp16 var_549_to_fp16 = const()[name = string("op_549_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_549_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; - tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16242752)))]; - tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16244864)))]; - fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_43_cast_fp16")]; - string var_569_pad_type_0 = const()[name = string("op_569_pad_type_0"), val = string("valid")]; - tensor var_569_strides_0 = const()[name = string("op_569_strides_0"), val = tensor([1, 1])]; - tensor var_569_pad_0 = const()[name = string("op_569_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_569_dilations_0 = const()[name = string("op_569_dilations_0"), val = tensor([1, 1])]; - int32 var_569_groups_0 = const()[name = string("op_569_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16246976))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17819904))))[name = string("layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_569_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_569_dilations_0, groups = var_569_groups_0, pad = var_569_pad_0, pad_type = var_569_pad_type_0, strides = var_569_strides_0, weight = layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = string("op_569_cast_fp16")]; - string var_575_pad_type_0 = const()[name = string("op_575_pad_type_0"), val = string("valid")]; - tensor var_575_strides_0 = const()[name = string("op_575_strides_0"), val = tensor([1, 1])]; - tensor var_575_pad_0 = const()[name = string("op_575_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_575_dilations_0 = const()[name = string("op_575_dilations_0"), val = tensor([1, 1])]; - int32 var_575_groups_0 = const()[name = string("op_575_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17874176))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17824064))))[name = string("layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_575_cast_fp16 = conv(dilations = var_575_dilations_0, groups = var_575_groups_0, pad = var_575_pad_0, pad_type = var_575_pad_type_0, strides = var_575_strides_0, weight = layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_43_cast_fp16)[name = string("op_575_cast_fp16")]; - tensor input_45_cast_fp16 = add(x = var_569_cast_fp16, y = var_575_cast_fp16)[name = string("input_45_cast_fp16")]; - tensor input_47_cast_fp16 = silu(x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; - string var_586_pad_type_0 = const()[name = string("op_586_pad_type_0"), val = string("valid")]; - tensor var_586_strides_0 = const()[name = string("op_586_strides_0"), val = tensor([1, 1])]; - tensor var_586_pad_0 = const()[name = string("op_586_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_586_dilations_0 = const()[name = string("op_586_dilations_0"), val = tensor([1, 1])]; - int32 var_586_groups_0 = const()[name = string("op_586_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18398528))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19971456))))[name = string("layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_586_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_586_dilations_0, groups = var_586_groups_0, pad = var_586_pad_0, pad_type = var_586_pad_type_0, strides = var_586_strides_0, weight = layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = string("op_586_cast_fp16")]; - string var_592_pad_type_0 = const()[name = string("op_592_pad_type_0"), val = string("valid")]; - tensor var_592_strides_0 = const()[name = string("op_592_strides_0"), val = tensor([1, 1])]; - tensor var_592_pad_0 = const()[name = string("op_592_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_592_dilations_0 = const()[name = string("op_592_dilations_0"), val = tensor([1, 1])]; - int32 var_592_groups_0 = const()[name = string("op_592_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20049408))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19972544))))[name = string("layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_592_cast_fp16 = conv(dilations = var_592_dilations_0, groups = var_592_groups_0, pad = var_592_pad_0, pad_type = var_592_pad_type_0, strides = var_592_strides_0, weight = layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_47_cast_fp16)[name = string("op_592_cast_fp16")]; - tensor x_9_cast_fp16 = add(x = var_586_cast_fp16, y = var_592_cast_fp16)[name = string("x_9_cast_fp16")]; - fp16 var_594_to_fp16 = const()[name = string("op_594_to_fp16"), val = fp16(0x1p-1)]; - tensor var_595_cast_fp16 = mul(x = x_9_cast_fp16, y = var_594_to_fp16)[name = string("op_595_cast_fp16")]; - tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = var_595_cast_fp16)[name = string("inputs_13_cast_fp16")]; - tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; - fp16 var_605_to_fp16 = const()[name = string("op_605_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_605_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; - tensor obj_7_gamma_0_to_fp16 = const()[name = string("obj_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20573760)))]; - tensor obj_7_beta_0_to_fp16 = const()[name = string("obj_7_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20575872)))]; - fp16 obj_7_epsilon_0_to_fp16 = const()[name = string("obj_7_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_7_cast_fp16 = batch_norm(beta = obj_7_beta_0_to_fp16, epsilon = obj_7_epsilon_0_to_fp16, gamma = obj_7_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_7_cast_fp16")]; - string var_630_pad_type_0 = const()[name = string("op_630_pad_type_0"), val = string("valid")]; - tensor var_630_strides_0 = const()[name = string("op_630_strides_0"), val = tensor([1, 1])]; - tensor var_630_pad_0 = const()[name = string("op_630_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_630_dilations_0 = const()[name = string("op_630_dilations_0"), val = tensor([1, 1])]; - int32 var_630_groups_0 = const()[name = string("op_630_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20577984))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20971264))))[name = string("layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_630_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_630_dilations_0, groups = var_630_groups_0, pad = var_630_pad_0, pad_type = var_630_pad_type_0, strides = var_630_strides_0, weight = layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = string("op_630_cast_fp16")]; - string var_636_pad_type_0 = const()[name = string("op_636_pad_type_0"), val = string("valid")]; - tensor var_636_strides_0 = const()[name = string("op_636_strides_0"), val = tensor([1, 1])]; - tensor var_636_pad_0 = const()[name = string("op_636_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_636_dilations_0 = const()[name = string("op_636_dilations_0"), val = tensor([1, 1])]; - int32 var_636_groups_0 = const()[name = string("op_636_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20985664))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20972352))))[name = string("layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_636_cast_fp16 = conv(dilations = var_636_dilations_0, groups = var_636_groups_0, pad = var_636_pad_0, pad_type = var_636_pad_type_0, strides = var_636_strides_0, weight = layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = string("op_636_cast_fp16")]; - tensor query_5_cast_fp16 = add(x = var_630_cast_fp16, y = var_636_cast_fp16)[name = string("query_5_cast_fp16")]; - string var_645_pad_type_0 = const()[name = string("op_645_pad_type_0"), val = string("valid")]; - tensor var_645_strides_0 = const()[name = string("op_645_strides_0"), val = tensor([1, 1])]; - tensor var_645_pad_0 = const()[name = string("op_645_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_645_dilations_0 = const()[name = string("op_645_dilations_0"), val = tensor([1, 1])]; - int32 var_645_groups_0 = const()[name = string("op_645_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21116800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21510080))))[name = string("layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_645_cast_fp16 = conv(dilations = var_645_dilations_0, groups = var_645_groups_0, pad = var_645_pad_0, pad_type = var_645_pad_type_0, strides = var_645_strides_0, weight = layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = string("op_645_cast_fp16")]; - string var_651_pad_type_0 = const()[name = string("op_651_pad_type_0"), val = string("valid")]; - tensor var_651_strides_0 = const()[name = string("op_651_strides_0"), val = tensor([1, 1])]; - tensor var_651_pad_0 = const()[name = string("op_651_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_651_dilations_0 = const()[name = string("op_651_dilations_0"), val = tensor([1, 1])]; - int32 var_651_groups_0 = const()[name = string("op_651_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21529536))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21511168))))[name = string("layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_651_cast_fp16 = conv(dilations = var_651_dilations_0, groups = var_651_groups_0, pad = var_651_pad_0, pad_type = var_651_pad_type_0, strides = var_651_strides_0, weight = layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = string("op_651_cast_fp16")]; - tensor key_3_cast_fp16 = add(x = var_645_cast_fp16, y = var_651_cast_fp16)[name = string("key_3_cast_fp16")]; - string var_661_pad_type_0 = const()[name = string("op_661_pad_type_0"), val = string("valid")]; - tensor var_661_strides_0 = const()[name = string("op_661_strides_0"), val = tensor([1, 1])]; - tensor var_661_pad_0 = const()[name = string("op_661_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_661_dilations_0 = const()[name = string("op_661_dilations_0"), val = tensor([1, 1])]; - int32 var_661_groups_0 = const()[name = string("op_661_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21660672))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22053952))))[name = string("layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_661_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_661_dilations_0, groups = var_661_groups_0, pad = var_661_pad_0, pad_type = var_661_pad_type_0, strides = var_661_strides_0, weight = layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = string("op_661_cast_fp16")]; - string var_667_pad_type_0 = const()[name = string("op_667_pad_type_0"), val = string("valid")]; - tensor var_667_strides_0 = const()[name = string("op_667_strides_0"), val = tensor([1, 1])]; - tensor var_667_pad_0 = const()[name = string("op_667_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_667_dilations_0 = const()[name = string("op_667_dilations_0"), val = tensor([1, 1])]; - int32 var_667_groups_0 = const()[name = string("op_667_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22064640))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22055040))))[name = string("layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_667_cast_fp16 = conv(dilations = var_667_dilations_0, groups = var_667_groups_0, pad = var_667_pad_0, pad_type = var_667_pad_type_0, strides = var_667_strides_0, weight = layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = string("op_667_cast_fp16")]; - tensor value_3_cast_fp16 = add(x = var_661_cast_fp16, y = var_667_cast_fp16)[name = string("value_3_cast_fp16")]; - tensor var_670_to_fp16 = const()[name = string("op_670_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22195776)))]; - tensor query_7_cast_fp16 = add(x = query_5_cast_fp16, y = var_670_to_fp16)[name = string("query_7_cast_fp16")]; - tensor var_673_to_fp16 = const()[name = string("op_673_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22197888)))]; - tensor q_with_bias_v_3_cast_fp16 = add(x = query_5_cast_fp16, y = var_673_to_fp16)[name = string("q_with_bias_v_3_cast_fp16")]; - string var_683_pad_type_0 = const()[name = string("op_683_pad_type_0"), val = string("valid")]; - tensor var_683_strides_0 = const()[name = string("op_683_strides_0"), val = tensor([1, 1])]; - tensor var_683_pad_0 = const()[name = string("op_683_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_683_dilations_0 = const()[name = string("op_683_dilations_0"), val = tensor([1, 1])]; - int32 var_683_groups_0 = const()[name = string("op_683_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22200000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22593280))))[name = string("layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_683_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_683_dilations_0, groups = var_683_groups_0, pad = var_683_pad_0, pad_type = var_683_pad_type_0, strides = var_683_strides_0, weight = layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_683_cast_fp16")]; - string var_689_pad_type_0 = const()[name = string("op_689_pad_type_0"), val = string("valid")]; - tensor var_689_strides_0 = const()[name = string("op_689_strides_0"), val = tensor([1, 1])]; - tensor var_689_pad_0 = const()[name = string("op_689_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_689_dilations_0 = const()[name = string("op_689_dilations_0"), val = tensor([1, 1])]; - int32 var_689_groups_0 = const()[name = string("op_689_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22630464))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22594368))))[name = string("layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_689_cast_fp16 = conv(dilations = var_689_dilations_0, groups = var_689_groups_0, pad = var_689_pad_0, pad_type = var_689_pad_type_0, strides = var_689_strides_0, weight = layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_689_cast_fp16")]; - tensor p_3_cast_fp16 = add(x = var_683_cast_fp16, y = var_689_cast_fp16)[name = string("p_3_cast_fp16")]; - tensor var_693 = const()[name = string("op_693"), val = tensor([1, 8, 128, 188])]; - tensor var_694_cast_fp16 = reshape(shape = var_693, x = q_with_bias_v_3_cast_fp16)[name = string("op_694_cast_fp16")]; - tensor var_695 = const()[name = string("op_695"), val = tensor([1, 8, 128, -1])]; - tensor var_696_cast_fp16 = reshape(shape = var_695, x = p_3_cast_fp16)[name = string("op_696_cast_fp16")]; - bool matrix_bd_9_transpose_x_0 = const()[name = string("matrix_bd_9_transpose_x_0"), val = bool(true)]; - bool matrix_bd_9_transpose_y_0 = const()[name = string("matrix_bd_9_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_9_cast_fp16 = matmul(transpose_x = matrix_bd_9_transpose_x_0, transpose_y = matrix_bd_9_transpose_y_0, x = var_694_cast_fp16, y = var_696_cast_fp16)[name = string("matrix_bd_9_cast_fp16")]; - tensor matrix_bd_11_pad_0 = const()[name = string("matrix_bd_11_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_11_mode_0 = const()[name = string("matrix_bd_11_mode_0"), val = string("constant")]; - fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_11_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = matrix_bd_11_mode_0, pad = matrix_bd_11_pad_0, x = matrix_bd_9_cast_fp16)[name = string("matrix_bd_11_cast_fp16")]; - tensor var_705 = const()[name = string("op_705"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_13_cast_fp16 = reshape(shape = var_705, x = matrix_bd_11_cast_fp16)[name = string("matrix_bd_13_cast_fp16")]; - tensor var_709_begin_0 = const()[name = string("op_709_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_709_end_0 = const()[name = string("op_709_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_709_end_mask_0 = const()[name = string("op_709_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_709_cast_fp16 = slice_by_index(begin = var_709_begin_0, end = var_709_end_0, end_mask = var_709_end_mask_0, x = matrix_bd_13_cast_fp16)[name = string("op_709_cast_fp16")]; - tensor var_710 = const()[name = string("op_710"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_15_cast_fp16 = reshape(shape = var_710, x = var_709_cast_fp16)[name = string("matrix_bd_15_cast_fp16")]; - tensor var_715_begin_0 = const()[name = string("op_715_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_715_end_0 = const()[name = string("op_715_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_715_end_mask_0 = const()[name = string("op_715_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_715_cast_fp16 = slice_by_index(begin = var_715_begin_0, end = var_715_end_0, end_mask = var_715_end_mask_0, x = matrix_bd_15_cast_fp16)[name = string("op_715_cast_fp16")]; - fp16 var_716_to_fp16 = const()[name = string("op_716_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_3_cast_fp16 = mul(x = var_715_cast_fp16, y = var_716_to_fp16)[name = string("qk_mask_3_cast_fp16")]; - tensor var_720 = const()[name = string("op_720"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_3_cast_fp16 = reshape(shape = var_720, x = query_7_cast_fp16)[name = string("mh_q_3_cast_fp16")]; - fp16 var_722_to_fp16 = const()[name = string("op_722_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_723_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_722_to_fp16)[name = string("op_723_cast_fp16")]; - tensor var_726 = const()[name = string("op_726"), val = tensor([1, 8, 128, 188])]; - tensor var_727_cast_fp16 = reshape(shape = var_726, x = key_3_cast_fp16)[name = string("op_727_cast_fp16")]; - bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; - bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; - tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_723_cast_fp16, y = var_727_cast_fp16)[name = string("mh_w_5_cast_fp16")]; - tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = qk_mask_3_cast_fp16)[name = string("mh_w_7_cast_fp16")]; - tensor var_731_cast_fp16 = softmax(axis = var_518, x = mh_w_7_cast_fp16)[name = string("op_731_cast_fp16")]; - tensor var_732 = const()[name = string("op_732"), val = tensor([1, 8, 128, 188])]; - tensor var_733_cast_fp16 = reshape(shape = var_732, x = value_3_cast_fp16)[name = string("op_733_cast_fp16")]; - bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; - bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; - tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_733_cast_fp16, y = var_731_cast_fp16)[name = string("attn_3_cast_fp16")]; - tensor var_736 = const()[name = string("op_736"), val = tensor([1, 1024, 1, 188])]; - tensor input_49_cast_fp16 = reshape(shape = var_736, x = attn_3_cast_fp16)[name = string("input_49_cast_fp16")]; - string var_746_pad_type_0 = const()[name = string("op_746_pad_type_0"), val = string("valid")]; - tensor var_746_strides_0 = const()[name = string("op_746_strides_0"), val = tensor([1, 1])]; - tensor var_746_pad_0 = const()[name = string("op_746_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_746_dilations_0 = const()[name = string("op_746_dilations_0"), val = tensor([1, 1])]; - int32 var_746_groups_0 = const()[name = string("op_746_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22761600))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23154880))))[name = string("layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_746_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_746_dilations_0, groups = var_746_groups_0, pad = var_746_pad_0, pad_type = var_746_pad_type_0, strides = var_746_strides_0, weight = layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = string("op_746_cast_fp16")]; - string var_752_pad_type_0 = const()[name = string("op_752_pad_type_0"), val = string("valid")]; - tensor var_752_strides_0 = const()[name = string("op_752_strides_0"), val = tensor([1, 1])]; - tensor var_752_pad_0 = const()[name = string("op_752_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_752_dilations_0 = const()[name = string("op_752_dilations_0"), val = tensor([1, 1])]; - int32 var_752_groups_0 = const()[name = string("op_752_groups_0"), val = int32(1)]; - tensor layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23166656))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23155968))))[name = string("layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_752_cast_fp16 = conv(dilations = var_752_dilations_0, groups = var_752_groups_0, pad = var_752_pad_0, pad_type = var_752_pad_type_0, strides = var_752_strides_0, weight = layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_49_cast_fp16)[name = string("op_752_cast_fp16")]; - tensor obj_9_cast_fp16 = add(x = var_746_cast_fp16, y = var_752_cast_fp16)[name = string("obj_9_cast_fp16")]; - tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_9_cast_fp16)[name = string("inputs_15_cast_fp16")]; - tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; - fp16 var_763_to_fp16 = const()[name = string("op_763_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_763_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; - tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23297792)))]; - tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23299904)))]; - fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_51_cast_fp16")]; - string var_784_pad_type_0 = const()[name = string("op_784_pad_type_0"), val = string("valid")]; - tensor var_784_strides_0 = const()[name = string("op_784_strides_0"), val = tensor([1, 1])]; - tensor var_784_pad_0 = const()[name = string("op_784_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_784_dilations_0 = const()[name = string("op_784_dilations_0"), val = tensor([1, 1])]; - int32 var_784_groups_0 = const()[name = string("op_784_groups_0"), val = int32(1)]; - tensor layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23302016))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24088512))))[name = string("layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_784_cast_fp16 = conv(dilations = var_784_dilations_0, groups = var_784_groups_0, pad = var_784_pad_0, pad_type = var_784_pad_type_0, strides = var_784_strides_0, weight = layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = string("op_784_cast_fp16")]; - string var_790_pad_type_0 = const()[name = string("op_790_pad_type_0"), val = string("valid")]; - tensor var_790_strides_0 = const()[name = string("op_790_strides_0"), val = tensor([1, 1])]; - tensor var_790_pad_0 = const()[name = string("op_790_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_790_dilations_0 = const()[name = string("op_790_dilations_0"), val = tensor([1, 1])]; - int32 var_790_groups_0 = const()[name = string("op_790_groups_0"), val = int32(1)]; - tensor layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24113920))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24090624))))[name = string("layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_790_cast_fp16 = conv(dilations = var_790_dilations_0, groups = var_790_groups_0, pad = var_790_pad_0, pad_type = var_790_pad_type_0, strides = var_790_strides_0, weight = layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_51_cast_fp16)[name = string("op_790_cast_fp16")]; - tensor input_53_cast_fp16 = add(x = var_784_cast_fp16, y = var_790_cast_fp16)[name = string("input_53_cast_fp16")]; - int32 input_55_split_num_splits_0 = const()[name = string("input_55_split_num_splits_0"), val = int32(2)]; - int32 input_55_split_axis_0 = const()[name = string("input_55_split_axis_0"), val = int32(1)]; - tensor input_55_split_cast_fp16_0, tensor input_55_split_cast_fp16_1 = split(axis = input_55_split_axis_0, num_splits = input_55_split_num_splits_0, x = input_53_cast_fp16)[name = string("input_55_split_cast_fp16")]; - tensor input_55_split_1_sigmoid_cast_fp16 = sigmoid(x = input_55_split_cast_fp16_1)[name = string("input_55_split_1_sigmoid_cast_fp16")]; - tensor input_55_cast_fp16 = mul(x = input_55_split_cast_fp16_0, y = input_55_split_1_sigmoid_cast_fp16)[name = string("input_55_cast_fp16")]; - string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("custom")]; - tensor input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1024)]; - tensor input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor([1, 1])]; - tensor input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor([1, 1])]; - tensor const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24376128)))]; - tensor const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24394624)))]; - tensor input_59_cast_fp16 = conv(bias = const_271_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = const_270_to_fp16, x = input_55_cast_fp16)[name = string("input_59_cast_fp16")]; - tensor input_61_cast_fp16 = silu(x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; - string var_812_pad_type_0 = const()[name = string("op_812_pad_type_0"), val = string("valid")]; - tensor var_812_strides_0 = const()[name = string("op_812_strides_0"), val = tensor([1, 1])]; - tensor var_812_pad_0 = const()[name = string("op_812_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_812_dilations_0 = const()[name = string("op_812_dilations_0"), val = tensor([1, 1])]; - int32 var_812_groups_0 = const()[name = string("op_812_groups_0"), val = int32(1)]; - tensor layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24396736))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24790016))))[name = string("layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_812_cast_fp16 = conv(dilations = var_812_dilations_0, groups = var_812_groups_0, pad = var_812_pad_0, pad_type = var_812_pad_type_0, strides = var_812_strides_0, weight = layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = string("op_812_cast_fp16")]; - string var_818_pad_type_0 = const()[name = string("op_818_pad_type_0"), val = string("valid")]; - tensor var_818_strides_0 = const()[name = string("op_818_strides_0"), val = tensor([1, 1])]; - tensor var_818_pad_0 = const()[name = string("op_818_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_818_dilations_0 = const()[name = string("op_818_dilations_0"), val = tensor([1, 1])]; - int32 var_818_groups_0 = const()[name = string("op_818_groups_0"), val = int32(1)]; - tensor layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24802944))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24791104))))[name = string("layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_818_cast_fp16 = conv(dilations = var_818_dilations_0, groups = var_818_groups_0, pad = var_818_pad_0, pad_type = var_818_pad_type_0, strides = var_818_strides_0, weight = layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_61_cast_fp16)[name = string("op_818_cast_fp16")]; - tensor x_11_cast_fp16 = add(x = var_812_cast_fp16, y = var_818_cast_fp16)[name = string("x_11_cast_fp16")]; - tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = x_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; - tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; - fp16 var_829_to_fp16 = const()[name = string("op_829_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_829_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; - tensor input_63_gamma_0_to_fp16 = const()[name = string("input_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24934080)))]; - tensor input_63_beta_0_to_fp16 = const()[name = string("input_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24936192)))]; - fp16 input_63_epsilon_0_to_fp16 = const()[name = string("input_63_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_63_cast_fp16 = batch_norm(beta = input_63_beta_0_to_fp16, epsilon = input_63_epsilon_0_to_fp16, gamma = input_63_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_63_cast_fp16")]; - string var_849_pad_type_0 = const()[name = string("op_849_pad_type_0"), val = string("valid")]; - tensor var_849_strides_0 = const()[name = string("op_849_strides_0"), val = tensor([1, 1])]; - tensor var_849_pad_0 = const()[name = string("op_849_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_849_dilations_0 = const()[name = string("op_849_dilations_0"), val = tensor([1, 1])]; - int32 var_849_groups_0 = const()[name = string("op_849_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24938304))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26511232))))[name = string("layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_849_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_849_dilations_0, groups = var_849_groups_0, pad = var_849_pad_0, pad_type = var_849_pad_type_0, strides = var_849_strides_0, weight = layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = string("op_849_cast_fp16")]; - string var_855_pad_type_0 = const()[name = string("op_855_pad_type_0"), val = string("valid")]; - tensor var_855_strides_0 = const()[name = string("op_855_strides_0"), val = tensor([1, 1])]; - tensor var_855_pad_0 = const()[name = string("op_855_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_855_dilations_0 = const()[name = string("op_855_dilations_0"), val = tensor([1, 1])]; - int32 var_855_groups_0 = const()[name = string("op_855_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26566272))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26515392))))[name = string("layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_855_cast_fp16 = conv(dilations = var_855_dilations_0, groups = var_855_groups_0, pad = var_855_pad_0, pad_type = var_855_pad_type_0, strides = var_855_strides_0, weight = layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_63_cast_fp16)[name = string("op_855_cast_fp16")]; - tensor input_65_cast_fp16 = add(x = var_849_cast_fp16, y = var_855_cast_fp16)[name = string("input_65_cast_fp16")]; - tensor input_67_cast_fp16 = silu(x = input_65_cast_fp16)[name = string("input_67_cast_fp16")]; - string var_866_pad_type_0 = const()[name = string("op_866_pad_type_0"), val = string("valid")]; - tensor var_866_strides_0 = const()[name = string("op_866_strides_0"), val = tensor([1, 1])]; - tensor var_866_pad_0 = const()[name = string("op_866_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_866_dilations_0 = const()[name = string("op_866_dilations_0"), val = tensor([1, 1])]; - int32 var_866_groups_0 = const()[name = string("op_866_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27090624))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28663552))))[name = string("layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_866_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_866_dilations_0, groups = var_866_groups_0, pad = var_866_pad_0, pad_type = var_866_pad_type_0, strides = var_866_strides_0, weight = layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = string("op_866_cast_fp16")]; - string var_872_pad_type_0 = const()[name = string("op_872_pad_type_0"), val = string("valid")]; - tensor var_872_strides_0 = const()[name = string("op_872_strides_0"), val = tensor([1, 1])]; - tensor var_872_pad_0 = const()[name = string("op_872_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_872_dilations_0 = const()[name = string("op_872_dilations_0"), val = tensor([1, 1])]; - int32 var_872_groups_0 = const()[name = string("op_872_groups_0"), val = int32(1)]; - tensor layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28735680))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28664640))))[name = string("layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_872_cast_fp16 = conv(dilations = var_872_dilations_0, groups = var_872_groups_0, pad = var_872_pad_0, pad_type = var_872_pad_type_0, strides = var_872_strides_0, weight = layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_67_cast_fp16)[name = string("op_872_cast_fp16")]; - tensor x_13_cast_fp16 = add(x = var_866_cast_fp16, y = var_872_cast_fp16)[name = string("x_13_cast_fp16")]; - fp16 var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = fp16(0x1p-1)]; - tensor var_875_cast_fp16 = mul(x = x_13_cast_fp16, y = var_874_to_fp16)[name = string("op_875_cast_fp16")]; - tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = var_875_cast_fp16)[name = string("inputs_19_cast_fp16")]; - tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; - fp16 var_885_to_fp16 = const()[name = string("op_885_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_885_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; - tensor inputs_21_gamma_0_to_fp16 = const()[name = string("inputs_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29260032)))]; - tensor inputs_21_beta_0_to_fp16 = const()[name = string("inputs_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29262144)))]; - fp16 inputs_21_epsilon_0_to_fp16 = const()[name = string("inputs_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_21_cast_fp16 = batch_norm(beta = inputs_21_beta_0_to_fp16, epsilon = inputs_21_epsilon_0_to_fp16, gamma = inputs_21_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("inputs_21_cast_fp16")]; - int32 var_899 = const()[name = string("op_899"), val = int32(3)]; - tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; - fp16 var_930_to_fp16 = const()[name = string("op_930_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_930_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; - tensor input_69_gamma_0_to_fp16 = const()[name = string("input_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29264256)))]; - tensor input_69_beta_0_to_fp16 = const()[name = string("input_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29266368)))]; - fp16 input_69_epsilon_0_to_fp16 = const()[name = string("input_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_69_cast_fp16 = batch_norm(beta = input_69_beta_0_to_fp16, epsilon = input_69_epsilon_0_to_fp16, gamma = input_69_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("input_69_cast_fp16")]; - string var_950_pad_type_0 = const()[name = string("op_950_pad_type_0"), val = string("valid")]; - tensor var_950_strides_0 = const()[name = string("op_950_strides_0"), val = tensor([1, 1])]; - tensor var_950_pad_0 = const()[name = string("op_950_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_950_dilations_0 = const()[name = string("op_950_dilations_0"), val = tensor([1, 1])]; - int32 var_950_groups_0 = const()[name = string("op_950_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29268480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30841408))))[name = string("layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_950_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_950_dilations_0, groups = var_950_groups_0, pad = var_950_pad_0, pad_type = var_950_pad_type_0, strides = var_950_strides_0, weight = layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = string("op_950_cast_fp16")]; - string var_956_pad_type_0 = const()[name = string("op_956_pad_type_0"), val = string("valid")]; - tensor var_956_strides_0 = const()[name = string("op_956_strides_0"), val = tensor([1, 1])]; - tensor var_956_pad_0 = const()[name = string("op_956_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_956_dilations_0 = const()[name = string("op_956_dilations_0"), val = tensor([1, 1])]; - int32 var_956_groups_0 = const()[name = string("op_956_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30884160))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30845568))))[name = string("layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_956_cast_fp16 = conv(dilations = var_956_dilations_0, groups = var_956_groups_0, pad = var_956_pad_0, pad_type = var_956_pad_type_0, strides = var_956_strides_0, weight = layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_69_cast_fp16)[name = string("op_956_cast_fp16")]; - tensor input_71_cast_fp16 = add(x = var_950_cast_fp16, y = var_956_cast_fp16)[name = string("input_71_cast_fp16")]; - tensor input_73_cast_fp16 = silu(x = input_71_cast_fp16)[name = string("input_73_cast_fp16")]; - string var_967_pad_type_0 = const()[name = string("op_967_pad_type_0"), val = string("valid")]; - tensor var_967_strides_0 = const()[name = string("op_967_strides_0"), val = tensor([1, 1])]; - tensor var_967_pad_0 = const()[name = string("op_967_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_967_dilations_0 = const()[name = string("op_967_dilations_0"), val = tensor([1, 1])]; - int32 var_967_groups_0 = const()[name = string("op_967_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31408512))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32981440))))[name = string("layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_967_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_967_dilations_0, groups = var_967_groups_0, pad = var_967_pad_0, pad_type = var_967_pad_type_0, strides = var_967_strides_0, weight = layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = string("op_967_cast_fp16")]; - string var_973_pad_type_0 = const()[name = string("op_973_pad_type_0"), val = string("valid")]; - tensor var_973_strides_0 = const()[name = string("op_973_strides_0"), val = tensor([1, 1])]; - tensor var_973_pad_0 = const()[name = string("op_973_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_973_dilations_0 = const()[name = string("op_973_dilations_0"), val = tensor([1, 1])]; - int32 var_973_groups_0 = const()[name = string("op_973_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33046400))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32982528))))[name = string("layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_973_cast_fp16 = conv(dilations = var_973_dilations_0, groups = var_973_groups_0, pad = var_973_pad_0, pad_type = var_973_pad_type_0, strides = var_973_strides_0, weight = layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_73_cast_fp16)[name = string("op_973_cast_fp16")]; - tensor x_15_cast_fp16 = add(x = var_967_cast_fp16, y = var_973_cast_fp16)[name = string("x_15_cast_fp16")]; - fp16 var_975_to_fp16 = const()[name = string("op_975_to_fp16"), val = fp16(0x1p-1)]; - tensor var_976_cast_fp16 = mul(x = x_15_cast_fp16, y = var_975_to_fp16)[name = string("op_976_cast_fp16")]; - tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = var_976_cast_fp16)[name = string("inputs_23_cast_fp16")]; - tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; - fp16 var_986_to_fp16 = const()[name = string("op_986_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_986_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; - tensor obj_11_gamma_0_to_fp16 = const()[name = string("obj_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33570752)))]; - tensor obj_11_beta_0_to_fp16 = const()[name = string("obj_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33572864)))]; - fp16 obj_11_epsilon_0_to_fp16 = const()[name = string("obj_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_11_cast_fp16 = batch_norm(beta = obj_11_beta_0_to_fp16, epsilon = obj_11_epsilon_0_to_fp16, gamma = obj_11_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("obj_11_cast_fp16")]; - string var_1011_pad_type_0 = const()[name = string("op_1011_pad_type_0"), val = string("valid")]; - tensor var_1011_strides_0 = const()[name = string("op_1011_strides_0"), val = tensor([1, 1])]; - tensor var_1011_pad_0 = const()[name = string("op_1011_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1011_dilations_0 = const()[name = string("op_1011_dilations_0"), val = tensor([1, 1])]; - int32 var_1011_groups_0 = const()[name = string("op_1011_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33574976))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33968256))))[name = string("layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1011_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1011_dilations_0, groups = var_1011_groups_0, pad = var_1011_pad_0, pad_type = var_1011_pad_type_0, strides = var_1011_strides_0, weight = layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = string("op_1011_cast_fp16")]; - string var_1017_pad_type_0 = const()[name = string("op_1017_pad_type_0"), val = string("valid")]; - tensor var_1017_strides_0 = const()[name = string("op_1017_strides_0"), val = tensor([1, 1])]; - tensor var_1017_pad_0 = const()[name = string("op_1017_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1017_dilations_0 = const()[name = string("op_1017_dilations_0"), val = tensor([1, 1])]; - int32 var_1017_groups_0 = const()[name = string("op_1017_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33983040))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33969344))))[name = string("layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1017_cast_fp16 = conv(dilations = var_1017_dilations_0, groups = var_1017_groups_0, pad = var_1017_pad_0, pad_type = var_1017_pad_type_0, strides = var_1017_strides_0, weight = layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = string("op_1017_cast_fp16")]; - tensor query_9_cast_fp16 = add(x = var_1011_cast_fp16, y = var_1017_cast_fp16)[name = string("query_9_cast_fp16")]; - string var_1026_pad_type_0 = const()[name = string("op_1026_pad_type_0"), val = string("valid")]; - tensor var_1026_strides_0 = const()[name = string("op_1026_strides_0"), val = tensor([1, 1])]; - tensor var_1026_pad_0 = const()[name = string("op_1026_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1026_dilations_0 = const()[name = string("op_1026_dilations_0"), val = tensor([1, 1])]; - int32 var_1026_groups_0 = const()[name = string("op_1026_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34114176))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34507456))))[name = string("layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1026_cast_fp16 = conv(dilations = var_1026_dilations_0, groups = var_1026_groups_0, pad = var_1026_pad_0, pad_type = var_1026_pad_type_0, strides = var_1026_strides_0, weight = layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = string("op_1026_cast_fp16")]; - string var_1032_pad_type_0 = const()[name = string("op_1032_pad_type_0"), val = string("valid")]; - tensor var_1032_strides_0 = const()[name = string("op_1032_strides_0"), val = tensor([1, 1])]; - tensor var_1032_pad_0 = const()[name = string("op_1032_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1032_dilations_0 = const()[name = string("op_1032_dilations_0"), val = tensor([1, 1])]; - int32 var_1032_groups_0 = const()[name = string("op_1032_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34523200))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34508544))))[name = string("layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1032_cast_fp16 = conv(dilations = var_1032_dilations_0, groups = var_1032_groups_0, pad = var_1032_pad_0, pad_type = var_1032_pad_type_0, strides = var_1032_strides_0, weight = layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = string("op_1032_cast_fp16")]; - tensor key_5_cast_fp16 = add(x = var_1026_cast_fp16, y = var_1032_cast_fp16)[name = string("key_5_cast_fp16")]; - string var_1042_pad_type_0 = const()[name = string("op_1042_pad_type_0"), val = string("valid")]; - tensor var_1042_strides_0 = const()[name = string("op_1042_strides_0"), val = tensor([1, 1])]; - tensor var_1042_pad_0 = const()[name = string("op_1042_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1042_dilations_0 = const()[name = string("op_1042_dilations_0"), val = tensor([1, 1])]; - int32 var_1042_groups_0 = const()[name = string("op_1042_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34654336))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35047616))))[name = string("layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1042_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1042_dilations_0, groups = var_1042_groups_0, pad = var_1042_pad_0, pad_type = var_1042_pad_type_0, strides = var_1042_strides_0, weight = layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = string("op_1042_cast_fp16")]; - string var_1048_pad_type_0 = const()[name = string("op_1048_pad_type_0"), val = string("valid")]; - tensor var_1048_strides_0 = const()[name = string("op_1048_strides_0"), val = tensor([1, 1])]; - tensor var_1048_pad_0 = const()[name = string("op_1048_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1048_dilations_0 = const()[name = string("op_1048_dilations_0"), val = tensor([1, 1])]; - int32 var_1048_groups_0 = const()[name = string("op_1048_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35056640))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35048704))))[name = string("layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1048_cast_fp16 = conv(dilations = var_1048_dilations_0, groups = var_1048_groups_0, pad = var_1048_pad_0, pad_type = var_1048_pad_type_0, strides = var_1048_strides_0, weight = layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = string("op_1048_cast_fp16")]; - tensor value_5_cast_fp16 = add(x = var_1042_cast_fp16, y = var_1048_cast_fp16)[name = string("value_5_cast_fp16")]; - tensor var_1051_to_fp16 = const()[name = string("op_1051_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35187776)))]; - tensor query_11_cast_fp16 = add(x = query_9_cast_fp16, y = var_1051_to_fp16)[name = string("query_11_cast_fp16")]; - tensor var_1054_to_fp16 = const()[name = string("op_1054_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35189888)))]; - tensor q_with_bias_v_5_cast_fp16 = add(x = query_9_cast_fp16, y = var_1054_to_fp16)[name = string("q_with_bias_v_5_cast_fp16")]; - string var_1064_pad_type_0 = const()[name = string("op_1064_pad_type_0"), val = string("valid")]; - tensor var_1064_strides_0 = const()[name = string("op_1064_strides_0"), val = tensor([1, 1])]; - tensor var_1064_pad_0 = const()[name = string("op_1064_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1064_dilations_0 = const()[name = string("op_1064_dilations_0"), val = tensor([1, 1])]; - int32 var_1064_groups_0 = const()[name = string("op_1064_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35192000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35585280))))[name = string("layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_1064_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1064_dilations_0, groups = var_1064_groups_0, pad = var_1064_pad_0, pad_type = var_1064_pad_type_0, strides = var_1064_strides_0, weight = layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_1064_cast_fp16")]; - string var_1070_pad_type_0 = const()[name = string("op_1070_pad_type_0"), val = string("valid")]; - tensor var_1070_strides_0 = const()[name = string("op_1070_strides_0"), val = tensor([1, 1])]; - tensor var_1070_pad_0 = const()[name = string("op_1070_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1070_dilations_0 = const()[name = string("op_1070_dilations_0"), val = tensor([1, 1])]; - int32 var_1070_groups_0 = const()[name = string("op_1070_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35623680))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35586368))))[name = string("layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1070_cast_fp16 = conv(dilations = var_1070_dilations_0, groups = var_1070_groups_0, pad = var_1070_pad_0, pad_type = var_1070_pad_type_0, strides = var_1070_strides_0, weight = layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_1070_cast_fp16")]; - tensor p_5_cast_fp16 = add(x = var_1064_cast_fp16, y = var_1070_cast_fp16)[name = string("p_5_cast_fp16")]; - tensor var_1074 = const()[name = string("op_1074"), val = tensor([1, 8, 128, 188])]; - tensor var_1075_cast_fp16 = reshape(shape = var_1074, x = q_with_bias_v_5_cast_fp16)[name = string("op_1075_cast_fp16")]; - tensor var_1076 = const()[name = string("op_1076"), val = tensor([1, 8, 128, -1])]; - tensor var_1077_cast_fp16 = reshape(shape = var_1076, x = p_5_cast_fp16)[name = string("op_1077_cast_fp16")]; - bool matrix_bd_17_transpose_x_0 = const()[name = string("matrix_bd_17_transpose_x_0"), val = bool(true)]; - bool matrix_bd_17_transpose_y_0 = const()[name = string("matrix_bd_17_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_17_cast_fp16 = matmul(transpose_x = matrix_bd_17_transpose_x_0, transpose_y = matrix_bd_17_transpose_y_0, x = var_1075_cast_fp16, y = var_1077_cast_fp16)[name = string("matrix_bd_17_cast_fp16")]; - tensor matrix_bd_19_pad_0 = const()[name = string("matrix_bd_19_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_19_mode_0 = const()[name = string("matrix_bd_19_mode_0"), val = string("constant")]; - fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_19_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = matrix_bd_19_mode_0, pad = matrix_bd_19_pad_0, x = matrix_bd_17_cast_fp16)[name = string("matrix_bd_19_cast_fp16")]; - tensor var_1086 = const()[name = string("op_1086"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_21_cast_fp16 = reshape(shape = var_1086, x = matrix_bd_19_cast_fp16)[name = string("matrix_bd_21_cast_fp16")]; - tensor var_1090_begin_0 = const()[name = string("op_1090_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_1090_end_0 = const()[name = string("op_1090_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_1090_end_mask_0 = const()[name = string("op_1090_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_1090_cast_fp16 = slice_by_index(begin = var_1090_begin_0, end = var_1090_end_0, end_mask = var_1090_end_mask_0, x = matrix_bd_21_cast_fp16)[name = string("op_1090_cast_fp16")]; - tensor var_1091 = const()[name = string("op_1091"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_23_cast_fp16 = reshape(shape = var_1091, x = var_1090_cast_fp16)[name = string("matrix_bd_23_cast_fp16")]; - tensor var_1096_begin_0 = const()[name = string("op_1096_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1096_end_0 = const()[name = string("op_1096_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_1096_end_mask_0 = const()[name = string("op_1096_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_1096_cast_fp16 = slice_by_index(begin = var_1096_begin_0, end = var_1096_end_0, end_mask = var_1096_end_mask_0, x = matrix_bd_23_cast_fp16)[name = string("op_1096_cast_fp16")]; - fp16 var_1097_to_fp16 = const()[name = string("op_1097_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_5_cast_fp16 = mul(x = var_1096_cast_fp16, y = var_1097_to_fp16)[name = string("qk_mask_5_cast_fp16")]; - tensor var_1101 = const()[name = string("op_1101"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_5_cast_fp16 = reshape(shape = var_1101, x = query_11_cast_fp16)[name = string("mh_q_5_cast_fp16")]; - fp16 var_1103_to_fp16 = const()[name = string("op_1103_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_1104_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_1103_to_fp16)[name = string("op_1104_cast_fp16")]; - tensor var_1107 = const()[name = string("op_1107"), val = tensor([1, 8, 128, 188])]; - tensor var_1108_cast_fp16 = reshape(shape = var_1107, x = key_5_cast_fp16)[name = string("op_1108_cast_fp16")]; - bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; - bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; - tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_1104_cast_fp16, y = var_1108_cast_fp16)[name = string("mh_w_9_cast_fp16")]; - tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = qk_mask_5_cast_fp16)[name = string("mh_w_11_cast_fp16")]; - tensor var_1112_cast_fp16 = softmax(axis = var_899, x = mh_w_11_cast_fp16)[name = string("op_1112_cast_fp16")]; - tensor var_1113 = const()[name = string("op_1113"), val = tensor([1, 8, 128, 188])]; - tensor var_1114_cast_fp16 = reshape(shape = var_1113, x = value_5_cast_fp16)[name = string("op_1114_cast_fp16")]; - bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; - bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; - tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_1114_cast_fp16, y = var_1112_cast_fp16)[name = string("attn_5_cast_fp16")]; - tensor var_1117 = const()[name = string("op_1117"), val = tensor([1, 1024, 1, 188])]; - tensor input_75_cast_fp16 = reshape(shape = var_1117, x = attn_5_cast_fp16)[name = string("input_75_cast_fp16")]; - string var_1127_pad_type_0 = const()[name = string("op_1127_pad_type_0"), val = string("valid")]; - tensor var_1127_strides_0 = const()[name = string("op_1127_strides_0"), val = tensor([1, 1])]; - tensor var_1127_pad_0 = const()[name = string("op_1127_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1127_dilations_0 = const()[name = string("op_1127_dilations_0"), val = tensor([1, 1])]; - int32 var_1127_groups_0 = const()[name = string("op_1127_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35754816))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36148096))))[name = string("layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1127_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1127_dilations_0, groups = var_1127_groups_0, pad = var_1127_pad_0, pad_type = var_1127_pad_type_0, strides = var_1127_strides_0, weight = layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = string("op_1127_cast_fp16")]; - string var_1133_pad_type_0 = const()[name = string("op_1133_pad_type_0"), val = string("valid")]; - tensor var_1133_strides_0 = const()[name = string("op_1133_strides_0"), val = tensor([1, 1])]; - tensor var_1133_pad_0 = const()[name = string("op_1133_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1133_dilations_0 = const()[name = string("op_1133_dilations_0"), val = tensor([1, 1])]; - int32 var_1133_groups_0 = const()[name = string("op_1133_groups_0"), val = int32(1)]; - tensor layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36158528))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36149184))))[name = string("layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1133_cast_fp16 = conv(dilations = var_1133_dilations_0, groups = var_1133_groups_0, pad = var_1133_pad_0, pad_type = var_1133_pad_type_0, strides = var_1133_strides_0, weight = layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_75_cast_fp16)[name = string("op_1133_cast_fp16")]; - tensor obj_13_cast_fp16 = add(x = var_1127_cast_fp16, y = var_1133_cast_fp16)[name = string("obj_13_cast_fp16")]; - tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = obj_13_cast_fp16)[name = string("inputs_25_cast_fp16")]; - tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; - fp16 var_1144_to_fp16 = const()[name = string("op_1144_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1144_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; - tensor input_77_gamma_0_to_fp16 = const()[name = string("input_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36289664)))]; - tensor input_77_beta_0_to_fp16 = const()[name = string("input_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36291776)))]; - fp16 input_77_epsilon_0_to_fp16 = const()[name = string("input_77_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_77_cast_fp16 = batch_norm(beta = input_77_beta_0_to_fp16, epsilon = input_77_epsilon_0_to_fp16, gamma = input_77_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("input_77_cast_fp16")]; - string var_1165_pad_type_0 = const()[name = string("op_1165_pad_type_0"), val = string("valid")]; - tensor var_1165_strides_0 = const()[name = string("op_1165_strides_0"), val = tensor([1, 1])]; - tensor var_1165_pad_0 = const()[name = string("op_1165_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1165_dilations_0 = const()[name = string("op_1165_dilations_0"), val = tensor([1, 1])]; - int32 var_1165_groups_0 = const()[name = string("op_1165_groups_0"), val = int32(1)]; - tensor layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36293888))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37080384))))[name = string("layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1165_cast_fp16 = conv(dilations = var_1165_dilations_0, groups = var_1165_groups_0, pad = var_1165_pad_0, pad_type = var_1165_pad_type_0, strides = var_1165_strides_0, weight = layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = string("op_1165_cast_fp16")]; - string var_1171_pad_type_0 = const()[name = string("op_1171_pad_type_0"), val = string("valid")]; - tensor var_1171_strides_0 = const()[name = string("op_1171_strides_0"), val = tensor([1, 1])]; - tensor var_1171_pad_0 = const()[name = string("op_1171_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1171_dilations_0 = const()[name = string("op_1171_dilations_0"), val = tensor([1, 1])]; - int32 var_1171_groups_0 = const()[name = string("op_1171_groups_0"), val = int32(1)]; - tensor layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37104064))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37082496))))[name = string("layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1171_cast_fp16 = conv(dilations = var_1171_dilations_0, groups = var_1171_groups_0, pad = var_1171_pad_0, pad_type = var_1171_pad_type_0, strides = var_1171_strides_0, weight = layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_77_cast_fp16)[name = string("op_1171_cast_fp16")]; - tensor input_79_cast_fp16 = add(x = var_1165_cast_fp16, y = var_1171_cast_fp16)[name = string("input_79_cast_fp16")]; - int32 input_81_split_num_splits_0 = const()[name = string("input_81_split_num_splits_0"), val = int32(2)]; - int32 input_81_split_axis_0 = const()[name = string("input_81_split_axis_0"), val = int32(1)]; - tensor input_81_split_cast_fp16_0, tensor input_81_split_cast_fp16_1 = split(axis = input_81_split_axis_0, num_splits = input_81_split_num_splits_0, x = input_79_cast_fp16)[name = string("input_81_split_cast_fp16")]; - tensor input_81_split_1_sigmoid_cast_fp16 = sigmoid(x = input_81_split_cast_fp16_1)[name = string("input_81_split_1_sigmoid_cast_fp16")]; - tensor input_81_cast_fp16 = mul(x = input_81_split_cast_fp16_0, y = input_81_split_1_sigmoid_cast_fp16)[name = string("input_81_cast_fp16")]; - string input_83_pad_type_0 = const()[name = string("input_83_pad_type_0"), val = string("custom")]; - tensor input_83_pad_0 = const()[name = string("input_83_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_83_groups_0 = const()[name = string("input_83_groups_0"), val = int32(1024)]; - tensor input_83_strides_0 = const()[name = string("input_83_strides_0"), val = tensor([1, 1])]; - tensor input_83_dilations_0 = const()[name = string("input_83_dilations_0"), val = tensor([1, 1])]; - tensor const_272_to_fp16 = const()[name = string("const_272_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37366272)))]; - tensor const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37384768)))]; - tensor input_85_cast_fp16 = conv(bias = const_273_to_fp16, dilations = input_83_dilations_0, groups = input_83_groups_0, pad = input_83_pad_0, pad_type = input_83_pad_type_0, strides = input_83_strides_0, weight = const_272_to_fp16, x = input_81_cast_fp16)[name = string("input_85_cast_fp16")]; - tensor input_87_cast_fp16 = silu(x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; - string var_1193_pad_type_0 = const()[name = string("op_1193_pad_type_0"), val = string("valid")]; - tensor var_1193_strides_0 = const()[name = string("op_1193_strides_0"), val = tensor([1, 1])]; - tensor var_1193_pad_0 = const()[name = string("op_1193_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1193_dilations_0 = const()[name = string("op_1193_dilations_0"), val = tensor([1, 1])]; - int32 var_1193_groups_0 = const()[name = string("op_1193_groups_0"), val = int32(1)]; - tensor layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37386880))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37780160))))[name = string("layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_1193_cast_fp16 = conv(dilations = var_1193_dilations_0, groups = var_1193_groups_0, pad = var_1193_pad_0, pad_type = var_1193_pad_type_0, strides = var_1193_strides_0, weight = layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = string("op_1193_cast_fp16")]; - string var_1199_pad_type_0 = const()[name = string("op_1199_pad_type_0"), val = string("valid")]; - tensor var_1199_strides_0 = const()[name = string("op_1199_strides_0"), val = tensor([1, 1])]; - tensor var_1199_pad_0 = const()[name = string("op_1199_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1199_dilations_0 = const()[name = string("op_1199_dilations_0"), val = tensor([1, 1])]; - int32 var_1199_groups_0 = const()[name = string("op_1199_groups_0"), val = int32(1)]; - tensor layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37791680))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37781248))))[name = string("layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1199_cast_fp16 = conv(dilations = var_1199_dilations_0, groups = var_1199_groups_0, pad = var_1199_pad_0, pad_type = var_1199_pad_type_0, strides = var_1199_strides_0, weight = layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_87_cast_fp16)[name = string("op_1199_cast_fp16")]; - tensor x_17_cast_fp16 = add(x = var_1193_cast_fp16, y = var_1199_cast_fp16)[name = string("x_17_cast_fp16")]; - tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = x_17_cast_fp16)[name = string("inputs_27_cast_fp16")]; - tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; - fp16 var_1210_to_fp16 = const()[name = string("op_1210_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1210_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; - tensor input_89_gamma_0_to_fp16 = const()[name = string("input_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37922816)))]; - tensor input_89_beta_0_to_fp16 = const()[name = string("input_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37924928)))]; - fp16 input_89_epsilon_0_to_fp16 = const()[name = string("input_89_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_89_cast_fp16 = batch_norm(beta = input_89_beta_0_to_fp16, epsilon = input_89_epsilon_0_to_fp16, gamma = input_89_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_89_cast_fp16")]; - string var_1230_pad_type_0 = const()[name = string("op_1230_pad_type_0"), val = string("valid")]; - tensor var_1230_strides_0 = const()[name = string("op_1230_strides_0"), val = tensor([1, 1])]; - tensor var_1230_pad_0 = const()[name = string("op_1230_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1230_dilations_0 = const()[name = string("op_1230_dilations_0"), val = tensor([1, 1])]; - int32 var_1230_groups_0 = const()[name = string("op_1230_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37927040))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39499968))))[name = string("layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1230_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1230_dilations_0, groups = var_1230_groups_0, pad = var_1230_pad_0, pad_type = var_1230_pad_type_0, strides = var_1230_strides_0, weight = layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = string("op_1230_cast_fp16")]; - string var_1236_pad_type_0 = const()[name = string("op_1236_pad_type_0"), val = string("valid")]; - tensor var_1236_strides_0 = const()[name = string("op_1236_strides_0"), val = tensor([1, 1])]; - tensor var_1236_pad_0 = const()[name = string("op_1236_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1236_dilations_0 = const()[name = string("op_1236_dilations_0"), val = tensor([1, 1])]; - int32 var_1236_groups_0 = const()[name = string("op_1236_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39552000))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39504128))))[name = string("layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1236_cast_fp16 = conv(dilations = var_1236_dilations_0, groups = var_1236_groups_0, pad = var_1236_pad_0, pad_type = var_1236_pad_type_0, strides = var_1236_strides_0, weight = layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_89_cast_fp16)[name = string("op_1236_cast_fp16")]; - tensor input_91_cast_fp16 = add(x = var_1230_cast_fp16, y = var_1236_cast_fp16)[name = string("input_91_cast_fp16")]; - tensor input_93_cast_fp16 = silu(x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; - string var_1247_pad_type_0 = const()[name = string("op_1247_pad_type_0"), val = string("valid")]; - tensor var_1247_strides_0 = const()[name = string("op_1247_strides_0"), val = tensor([1, 1])]; - tensor var_1247_pad_0 = const()[name = string("op_1247_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1247_dilations_0 = const()[name = string("op_1247_dilations_0"), val = tensor([1, 1])]; - int32 var_1247_groups_0 = const()[name = string("op_1247_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40076352))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41649280))))[name = string("layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_1247_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1247_dilations_0, groups = var_1247_groups_0, pad = var_1247_pad_0, pad_type = var_1247_pad_type_0, strides = var_1247_strides_0, weight = layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = string("op_1247_cast_fp16")]; - string var_1253_pad_type_0 = const()[name = string("op_1253_pad_type_0"), val = string("valid")]; - tensor var_1253_strides_0 = const()[name = string("op_1253_strides_0"), val = tensor([1, 1])]; - tensor var_1253_pad_0 = const()[name = string("op_1253_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1253_dilations_0 = const()[name = string("op_1253_dilations_0"), val = tensor([1, 1])]; - int32 var_1253_groups_0 = const()[name = string("op_1253_groups_0"), val = int32(1)]; - tensor layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41716736))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41650368))))[name = string("layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1253_cast_fp16 = conv(dilations = var_1253_dilations_0, groups = var_1253_groups_0, pad = var_1253_pad_0, pad_type = var_1253_pad_type_0, strides = var_1253_strides_0, weight = layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_93_cast_fp16)[name = string("op_1253_cast_fp16")]; - tensor x_19_cast_fp16 = add(x = var_1247_cast_fp16, y = var_1253_cast_fp16)[name = string("x_19_cast_fp16")]; - fp16 var_1255_to_fp16 = const()[name = string("op_1255_to_fp16"), val = fp16(0x1p-1)]; - tensor var_1256_cast_fp16 = mul(x = x_19_cast_fp16, y = var_1255_to_fp16)[name = string("op_1256_cast_fp16")]; - tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = var_1256_cast_fp16)[name = string("inputs_29_cast_fp16")]; - tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; - fp16 var_1266_to_fp16 = const()[name = string("op_1266_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1266_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; - tensor inputs_31_gamma_0_to_fp16 = const()[name = string("inputs_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42241088)))]; - tensor inputs_31_beta_0_to_fp16 = const()[name = string("inputs_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42243200)))]; - fp16 inputs_31_epsilon_0_to_fp16 = const()[name = string("inputs_31_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_31_cast_fp16 = batch_norm(beta = inputs_31_beta_0_to_fp16, epsilon = inputs_31_epsilon_0_to_fp16, gamma = inputs_31_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("inputs_31_cast_fp16")]; - int32 var_1280 = const()[name = string("op_1280"), val = int32(3)]; - tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; - fp16 var_1311_to_fp16 = const()[name = string("op_1311_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1311_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; - tensor input_95_gamma_0_to_fp16 = const()[name = string("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42245312)))]; - tensor input_95_beta_0_to_fp16 = const()[name = string("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42247424)))]; - fp16 input_95_epsilon_0_to_fp16 = const()[name = string("input_95_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_95_cast_fp16")]; - string var_1331_pad_type_0 = const()[name = string("op_1331_pad_type_0"), val = string("valid")]; - tensor var_1331_strides_0 = const()[name = string("op_1331_strides_0"), val = tensor([1, 1])]; - tensor var_1331_pad_0 = const()[name = string("op_1331_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1331_dilations_0 = const()[name = string("op_1331_dilations_0"), val = tensor([1, 1])]; - int32 var_1331_groups_0 = const()[name = string("op_1331_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42249536))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43822464))))[name = string("layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1331_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1331_dilations_0, groups = var_1331_groups_0, pad = var_1331_pad_0, pad_type = var_1331_pad_type_0, strides = var_1331_strides_0, weight = layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = string("op_1331_cast_fp16")]; - string var_1337_pad_type_0 = const()[name = string("op_1337_pad_type_0"), val = string("valid")]; - tensor var_1337_strides_0 = const()[name = string("op_1337_strides_0"), val = tensor([1, 1])]; - tensor var_1337_pad_0 = const()[name = string("op_1337_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1337_dilations_0 = const()[name = string("op_1337_dilations_0"), val = tensor([1, 1])]; - int32 var_1337_groups_0 = const()[name = string("op_1337_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43868416))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43826624))))[name = string("layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1337_cast_fp16 = conv(dilations = var_1337_dilations_0, groups = var_1337_groups_0, pad = var_1337_pad_0, pad_type = var_1337_pad_type_0, strides = var_1337_strides_0, weight = layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_95_cast_fp16)[name = string("op_1337_cast_fp16")]; - tensor input_97_cast_fp16 = add(x = var_1331_cast_fp16, y = var_1337_cast_fp16)[name = string("input_97_cast_fp16")]; - tensor input_99_cast_fp16 = silu(x = input_97_cast_fp16)[name = string("input_99_cast_fp16")]; - string var_1348_pad_type_0 = const()[name = string("op_1348_pad_type_0"), val = string("valid")]; - tensor var_1348_strides_0 = const()[name = string("op_1348_strides_0"), val = tensor([1, 1])]; - tensor var_1348_pad_0 = const()[name = string("op_1348_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1348_dilations_0 = const()[name = string("op_1348_dilations_0"), val = tensor([1, 1])]; - int32 var_1348_groups_0 = const()[name = string("op_1348_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44392768))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45965696))))[name = string("layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_1348_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1348_dilations_0, groups = var_1348_groups_0, pad = var_1348_pad_0, pad_type = var_1348_pad_type_0, strides = var_1348_strides_0, weight = layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = string("op_1348_cast_fp16")]; - string var_1354_pad_type_0 = const()[name = string("op_1354_pad_type_0"), val = string("valid")]; - tensor var_1354_strides_0 = const()[name = string("op_1354_strides_0"), val = tensor([1, 1])]; - tensor var_1354_pad_0 = const()[name = string("op_1354_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1354_dilations_0 = const()[name = string("op_1354_dilations_0"), val = tensor([1, 1])]; - int32 var_1354_groups_0 = const()[name = string("op_1354_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46042432))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45966784))))[name = string("layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1354_cast_fp16 = conv(dilations = var_1354_dilations_0, groups = var_1354_groups_0, pad = var_1354_pad_0, pad_type = var_1354_pad_type_0, strides = var_1354_strides_0, weight = layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_99_cast_fp16)[name = string("op_1354_cast_fp16")]; - tensor x_21_cast_fp16 = add(x = var_1348_cast_fp16, y = var_1354_cast_fp16)[name = string("x_21_cast_fp16")]; - fp16 var_1356_to_fp16 = const()[name = string("op_1356_to_fp16"), val = fp16(0x1p-1)]; - tensor var_1357_cast_fp16 = mul(x = x_21_cast_fp16, y = var_1356_to_fp16)[name = string("op_1357_cast_fp16")]; - tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = var_1357_cast_fp16)[name = string("inputs_33_cast_fp16")]; - tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; - fp16 var_1367_to_fp16 = const()[name = string("op_1367_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1367_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; - tensor obj_15_gamma_0_to_fp16 = const()[name = string("obj_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46566784)))]; - tensor obj_15_beta_0_to_fp16 = const()[name = string("obj_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46568896)))]; - fp16 obj_15_epsilon_0_to_fp16 = const()[name = string("obj_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_15_cast_fp16")]; - string var_1392_pad_type_0 = const()[name = string("op_1392_pad_type_0"), val = string("valid")]; - tensor var_1392_strides_0 = const()[name = string("op_1392_strides_0"), val = tensor([1, 1])]; - tensor var_1392_pad_0 = const()[name = string("op_1392_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1392_dilations_0 = const()[name = string("op_1392_dilations_0"), val = tensor([1, 1])]; - int32 var_1392_groups_0 = const()[name = string("op_1392_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46571008))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46964288))))[name = string("layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1392_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1392_dilations_0, groups = var_1392_groups_0, pad = var_1392_pad_0, pad_type = var_1392_pad_type_0, strides = var_1392_strides_0, weight = layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = string("op_1392_cast_fp16")]; - string var_1398_pad_type_0 = const()[name = string("op_1398_pad_type_0"), val = string("valid")]; - tensor var_1398_strides_0 = const()[name = string("op_1398_strides_0"), val = tensor([1, 1])]; - tensor var_1398_pad_0 = const()[name = string("op_1398_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1398_dilations_0 = const()[name = string("op_1398_dilations_0"), val = tensor([1, 1])]; - int32 var_1398_groups_0 = const()[name = string("op_1398_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46978112))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46965376))))[name = string("layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1398_cast_fp16 = conv(dilations = var_1398_dilations_0, groups = var_1398_groups_0, pad = var_1398_pad_0, pad_type = var_1398_pad_type_0, strides = var_1398_strides_0, weight = layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = string("op_1398_cast_fp16")]; - tensor query_13_cast_fp16 = add(x = var_1392_cast_fp16, y = var_1398_cast_fp16)[name = string("query_13_cast_fp16")]; - string var_1407_pad_type_0 = const()[name = string("op_1407_pad_type_0"), val = string("valid")]; - tensor var_1407_strides_0 = const()[name = string("op_1407_strides_0"), val = tensor([1, 1])]; - tensor var_1407_pad_0 = const()[name = string("op_1407_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1407_dilations_0 = const()[name = string("op_1407_dilations_0"), val = tensor([1, 1])]; - int32 var_1407_groups_0 = const()[name = string("op_1407_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47109248))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47502528))))[name = string("layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1407_cast_fp16 = conv(dilations = var_1407_dilations_0, groups = var_1407_groups_0, pad = var_1407_pad_0, pad_type = var_1407_pad_type_0, strides = var_1407_strides_0, weight = layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = string("op_1407_cast_fp16")]; - string var_1413_pad_type_0 = const()[name = string("op_1413_pad_type_0"), val = string("valid")]; - tensor var_1413_strides_0 = const()[name = string("op_1413_strides_0"), val = tensor([1, 1])]; - tensor var_1413_pad_0 = const()[name = string("op_1413_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1413_dilations_0 = const()[name = string("op_1413_dilations_0"), val = tensor([1, 1])]; - int32 var_1413_groups_0 = const()[name = string("op_1413_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47516864))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47503616))))[name = string("layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1413_cast_fp16 = conv(dilations = var_1413_dilations_0, groups = var_1413_groups_0, pad = var_1413_pad_0, pad_type = var_1413_pad_type_0, strides = var_1413_strides_0, weight = layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = string("op_1413_cast_fp16")]; - tensor key_7_cast_fp16 = add(x = var_1407_cast_fp16, y = var_1413_cast_fp16)[name = string("key_7_cast_fp16")]; - string var_1423_pad_type_0 = const()[name = string("op_1423_pad_type_0"), val = string("valid")]; - tensor var_1423_strides_0 = const()[name = string("op_1423_strides_0"), val = tensor([1, 1])]; - tensor var_1423_pad_0 = const()[name = string("op_1423_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1423_dilations_0 = const()[name = string("op_1423_dilations_0"), val = tensor([1, 1])]; - int32 var_1423_groups_0 = const()[name = string("op_1423_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47648000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48041280))))[name = string("layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1423_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1423_dilations_0, groups = var_1423_groups_0, pad = var_1423_pad_0, pad_type = var_1423_pad_type_0, strides = var_1423_strides_0, weight = layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = string("op_1423_cast_fp16")]; - string var_1429_pad_type_0 = const()[name = string("op_1429_pad_type_0"), val = string("valid")]; - tensor var_1429_strides_0 = const()[name = string("op_1429_strides_0"), val = tensor([1, 1])]; - tensor var_1429_pad_0 = const()[name = string("op_1429_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1429_dilations_0 = const()[name = string("op_1429_dilations_0"), val = tensor([1, 1])]; - int32 var_1429_groups_0 = const()[name = string("op_1429_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48050624))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48042368))))[name = string("layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1429_cast_fp16 = conv(dilations = var_1429_dilations_0, groups = var_1429_groups_0, pad = var_1429_pad_0, pad_type = var_1429_pad_type_0, strides = var_1429_strides_0, weight = layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = string("op_1429_cast_fp16")]; - tensor value_7_cast_fp16 = add(x = var_1423_cast_fp16, y = var_1429_cast_fp16)[name = string("value_7_cast_fp16")]; - tensor var_1432_to_fp16 = const()[name = string("op_1432_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48181760)))]; - tensor query_15_cast_fp16 = add(x = query_13_cast_fp16, y = var_1432_to_fp16)[name = string("query_15_cast_fp16")]; - tensor var_1435_to_fp16 = const()[name = string("op_1435_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48183872)))]; - tensor q_with_bias_v_7_cast_fp16 = add(x = query_13_cast_fp16, y = var_1435_to_fp16)[name = string("q_with_bias_v_7_cast_fp16")]; - string var_1445_pad_type_0 = const()[name = string("op_1445_pad_type_0"), val = string("valid")]; - tensor var_1445_strides_0 = const()[name = string("op_1445_strides_0"), val = tensor([1, 1])]; - tensor var_1445_pad_0 = const()[name = string("op_1445_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1445_dilations_0 = const()[name = string("op_1445_dilations_0"), val = tensor([1, 1])]; - int32 var_1445_groups_0 = const()[name = string("op_1445_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48185984))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48579264))))[name = string("layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_1445_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1445_dilations_0, groups = var_1445_groups_0, pad = var_1445_pad_0, pad_type = var_1445_pad_type_0, strides = var_1445_strides_0, weight = layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_1445_cast_fp16")]; - string var_1451_pad_type_0 = const()[name = string("op_1451_pad_type_0"), val = string("valid")]; - tensor var_1451_strides_0 = const()[name = string("op_1451_strides_0"), val = tensor([1, 1])]; - tensor var_1451_pad_0 = const()[name = string("op_1451_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1451_dilations_0 = const()[name = string("op_1451_dilations_0"), val = tensor([1, 1])]; - int32 var_1451_groups_0 = const()[name = string("op_1451_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48621056))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48580352))))[name = string("layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1451_cast_fp16 = conv(dilations = var_1451_dilations_0, groups = var_1451_groups_0, pad = var_1451_pad_0, pad_type = var_1451_pad_type_0, strides = var_1451_strides_0, weight = layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_1451_cast_fp16")]; - tensor p_7_cast_fp16 = add(x = var_1445_cast_fp16, y = var_1451_cast_fp16)[name = string("p_7_cast_fp16")]; - tensor var_1455 = const()[name = string("op_1455"), val = tensor([1, 8, 128, 188])]; - tensor var_1456_cast_fp16 = reshape(shape = var_1455, x = q_with_bias_v_7_cast_fp16)[name = string("op_1456_cast_fp16")]; - tensor var_1457 = const()[name = string("op_1457"), val = tensor([1, 8, 128, -1])]; - tensor var_1458_cast_fp16 = reshape(shape = var_1457, x = p_7_cast_fp16)[name = string("op_1458_cast_fp16")]; - bool matrix_bd_25_transpose_x_0 = const()[name = string("matrix_bd_25_transpose_x_0"), val = bool(true)]; - bool matrix_bd_25_transpose_y_0 = const()[name = string("matrix_bd_25_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_25_cast_fp16 = matmul(transpose_x = matrix_bd_25_transpose_x_0, transpose_y = matrix_bd_25_transpose_y_0, x = var_1456_cast_fp16, y = var_1458_cast_fp16)[name = string("matrix_bd_25_cast_fp16")]; - tensor matrix_bd_27_pad_0 = const()[name = string("matrix_bd_27_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_27_mode_0 = const()[name = string("matrix_bd_27_mode_0"), val = string("constant")]; - fp16 const_43_to_fp16 = const()[name = string("const_43_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_27_cast_fp16 = pad(constant_val = const_43_to_fp16, mode = matrix_bd_27_mode_0, pad = matrix_bd_27_pad_0, x = matrix_bd_25_cast_fp16)[name = string("matrix_bd_27_cast_fp16")]; - tensor var_1467 = const()[name = string("op_1467"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_29_cast_fp16 = reshape(shape = var_1467, x = matrix_bd_27_cast_fp16)[name = string("matrix_bd_29_cast_fp16")]; - tensor var_1471_begin_0 = const()[name = string("op_1471_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_1471_end_0 = const()[name = string("op_1471_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_1471_end_mask_0 = const()[name = string("op_1471_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_1471_cast_fp16 = slice_by_index(begin = var_1471_begin_0, end = var_1471_end_0, end_mask = var_1471_end_mask_0, x = matrix_bd_29_cast_fp16)[name = string("op_1471_cast_fp16")]; - tensor var_1472 = const()[name = string("op_1472"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_31_cast_fp16 = reshape(shape = var_1472, x = var_1471_cast_fp16)[name = string("matrix_bd_31_cast_fp16")]; - tensor var_1477_begin_0 = const()[name = string("op_1477_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1477_end_0 = const()[name = string("op_1477_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_1477_end_mask_0 = const()[name = string("op_1477_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_1477_cast_fp16 = slice_by_index(begin = var_1477_begin_0, end = var_1477_end_0, end_mask = var_1477_end_mask_0, x = matrix_bd_31_cast_fp16)[name = string("op_1477_cast_fp16")]; - fp16 var_1478_to_fp16 = const()[name = string("op_1478_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_7_cast_fp16 = mul(x = var_1477_cast_fp16, y = var_1478_to_fp16)[name = string("qk_mask_7_cast_fp16")]; - tensor var_1482 = const()[name = string("op_1482"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_7_cast_fp16 = reshape(shape = var_1482, x = query_15_cast_fp16)[name = string("mh_q_7_cast_fp16")]; - fp16 var_1484_to_fp16 = const()[name = string("op_1484_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_1485_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_1484_to_fp16)[name = string("op_1485_cast_fp16")]; - tensor var_1488 = const()[name = string("op_1488"), val = tensor([1, 8, 128, 188])]; - tensor var_1489_cast_fp16 = reshape(shape = var_1488, x = key_7_cast_fp16)[name = string("op_1489_cast_fp16")]; - bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; - bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; - tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_1485_cast_fp16, y = var_1489_cast_fp16)[name = string("mh_w_13_cast_fp16")]; - tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = qk_mask_7_cast_fp16)[name = string("mh_w_15_cast_fp16")]; - tensor var_1493_cast_fp16 = softmax(axis = var_1280, x = mh_w_15_cast_fp16)[name = string("op_1493_cast_fp16")]; - tensor var_1494 = const()[name = string("op_1494"), val = tensor([1, 8, 128, 188])]; - tensor var_1495_cast_fp16 = reshape(shape = var_1494, x = value_7_cast_fp16)[name = string("op_1495_cast_fp16")]; - bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; - bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; - tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_1495_cast_fp16, y = var_1493_cast_fp16)[name = string("attn_7_cast_fp16")]; - tensor var_1498 = const()[name = string("op_1498"), val = tensor([1, 1024, 1, 188])]; - tensor input_101_cast_fp16 = reshape(shape = var_1498, x = attn_7_cast_fp16)[name = string("input_101_cast_fp16")]; - string var_1508_pad_type_0 = const()[name = string("op_1508_pad_type_0"), val = string("valid")]; - tensor var_1508_strides_0 = const()[name = string("op_1508_strides_0"), val = tensor([1, 1])]; - tensor var_1508_pad_0 = const()[name = string("op_1508_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1508_dilations_0 = const()[name = string("op_1508_dilations_0"), val = tensor([1, 1])]; - int32 var_1508_groups_0 = const()[name = string("op_1508_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48752192))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49145472))))[name = string("layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1508_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1508_dilations_0, groups = var_1508_groups_0, pad = var_1508_pad_0, pad_type = var_1508_pad_type_0, strides = var_1508_strides_0, weight = layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_101_cast_fp16)[name = string("op_1508_cast_fp16")]; - string var_1514_pad_type_0 = const()[name = string("op_1514_pad_type_0"), val = string("valid")]; - tensor var_1514_strides_0 = const()[name = string("op_1514_strides_0"), val = tensor([1, 1])]; - tensor var_1514_pad_0 = const()[name = string("op_1514_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1514_dilations_0 = const()[name = string("op_1514_dilations_0"), val = tensor([1, 1])]; - int32 var_1514_groups_0 = const()[name = string("op_1514_groups_0"), val = int32(1)]; - tensor layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49155584))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49146560))))[name = string("layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1514_cast_fp16 = conv(dilations = var_1514_dilations_0, groups = var_1514_groups_0, pad = var_1514_pad_0, pad_type = var_1514_pad_type_0, strides = var_1514_strides_0, weight = layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_101_cast_fp16)[name = string("op_1514_cast_fp16")]; - tensor obj_17_cast_fp16 = add(x = var_1508_cast_fp16, y = var_1514_cast_fp16)[name = string("obj_17_cast_fp16")]; - tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_17_cast_fp16)[name = string("inputs_35_cast_fp16")]; - tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; - fp16 var_1525_to_fp16 = const()[name = string("op_1525_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1525_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; - tensor input_103_gamma_0_to_fp16 = const()[name = string("input_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49286720)))]; - tensor input_103_beta_0_to_fp16 = const()[name = string("input_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49288832)))]; - fp16 input_103_epsilon_0_to_fp16 = const()[name = string("input_103_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_103_cast_fp16 = batch_norm(beta = input_103_beta_0_to_fp16, epsilon = input_103_epsilon_0_to_fp16, gamma = input_103_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_103_cast_fp16")]; - string var_1546_pad_type_0 = const()[name = string("op_1546_pad_type_0"), val = string("valid")]; - tensor var_1546_strides_0 = const()[name = string("op_1546_strides_0"), val = tensor([1, 1])]; - tensor var_1546_pad_0 = const()[name = string("op_1546_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1546_dilations_0 = const()[name = string("op_1546_dilations_0"), val = tensor([1, 1])]; - int32 var_1546_groups_0 = const()[name = string("op_1546_groups_0"), val = int32(1)]; - tensor layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49290944))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50077440))))[name = string("layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1546_cast_fp16 = conv(dilations = var_1546_dilations_0, groups = var_1546_groups_0, pad = var_1546_pad_0, pad_type = var_1546_pad_type_0, strides = var_1546_strides_0, weight = layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = string("op_1546_cast_fp16")]; - string var_1552_pad_type_0 = const()[name = string("op_1552_pad_type_0"), val = string("valid")]; - tensor var_1552_strides_0 = const()[name = string("op_1552_strides_0"), val = tensor([1, 1])]; - tensor var_1552_pad_0 = const()[name = string("op_1552_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1552_dilations_0 = const()[name = string("op_1552_dilations_0"), val = tensor([1, 1])]; - int32 var_1552_groups_0 = const()[name = string("op_1552_groups_0"), val = int32(1)]; - tensor layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50100416))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50079552))))[name = string("layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1552_cast_fp16 = conv(dilations = var_1552_dilations_0, groups = var_1552_groups_0, pad = var_1552_pad_0, pad_type = var_1552_pad_type_0, strides = var_1552_strides_0, weight = layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_103_cast_fp16)[name = string("op_1552_cast_fp16")]; - tensor input_105_cast_fp16 = add(x = var_1546_cast_fp16, y = var_1552_cast_fp16)[name = string("input_105_cast_fp16")]; - int32 input_107_split_num_splits_0 = const()[name = string("input_107_split_num_splits_0"), val = int32(2)]; - int32 input_107_split_axis_0 = const()[name = string("input_107_split_axis_0"), val = int32(1)]; - tensor input_107_split_cast_fp16_0, tensor input_107_split_cast_fp16_1 = split(axis = input_107_split_axis_0, num_splits = input_107_split_num_splits_0, x = input_105_cast_fp16)[name = string("input_107_split_cast_fp16")]; - tensor input_107_split_1_sigmoid_cast_fp16 = sigmoid(x = input_107_split_cast_fp16_1)[name = string("input_107_split_1_sigmoid_cast_fp16")]; - tensor input_107_cast_fp16 = mul(x = input_107_split_cast_fp16_0, y = input_107_split_1_sigmoid_cast_fp16)[name = string("input_107_cast_fp16")]; - string input_109_pad_type_0 = const()[name = string("input_109_pad_type_0"), val = string("custom")]; - tensor input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_109_groups_0 = const()[name = string("input_109_groups_0"), val = int32(1024)]; - tensor input_109_strides_0 = const()[name = string("input_109_strides_0"), val = tensor([1, 1])]; - tensor input_109_dilations_0 = const()[name = string("input_109_dilations_0"), val = tensor([1, 1])]; - tensor const_274_to_fp16 = const()[name = string("const_274_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50362624)))]; - tensor const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50381120)))]; - tensor input_111_cast_fp16 = conv(bias = const_275_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = const_274_to_fp16, x = input_107_cast_fp16)[name = string("input_111_cast_fp16")]; - tensor input_113_cast_fp16 = silu(x = input_111_cast_fp16)[name = string("input_113_cast_fp16")]; - string var_1574_pad_type_0 = const()[name = string("op_1574_pad_type_0"), val = string("valid")]; - tensor var_1574_strides_0 = const()[name = string("op_1574_strides_0"), val = tensor([1, 1])]; - tensor var_1574_pad_0 = const()[name = string("op_1574_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1574_dilations_0 = const()[name = string("op_1574_dilations_0"), val = tensor([1, 1])]; - int32 var_1574_groups_0 = const()[name = string("op_1574_groups_0"), val = int32(1)]; - tensor layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50383232))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50776512))))[name = string("layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_1574_cast_fp16 = conv(dilations = var_1574_dilations_0, groups = var_1574_groups_0, pad = var_1574_pad_0, pad_type = var_1574_pad_type_0, strides = var_1574_strides_0, weight = layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = string("op_1574_cast_fp16")]; - string var_1580_pad_type_0 = const()[name = string("op_1580_pad_type_0"), val = string("valid")]; - tensor var_1580_strides_0 = const()[name = string("op_1580_strides_0"), val = tensor([1, 1])]; - tensor var_1580_pad_0 = const()[name = string("op_1580_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1580_dilations_0 = const()[name = string("op_1580_dilations_0"), val = tensor([1, 1])]; - int32 var_1580_groups_0 = const()[name = string("op_1580_groups_0"), val = int32(1)]; - tensor layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50788032))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50777600))))[name = string("layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1580_cast_fp16 = conv(dilations = var_1580_dilations_0, groups = var_1580_groups_0, pad = var_1580_pad_0, pad_type = var_1580_pad_type_0, strides = var_1580_strides_0, weight = layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_113_cast_fp16)[name = string("op_1580_cast_fp16")]; - tensor x_23_cast_fp16 = add(x = var_1574_cast_fp16, y = var_1580_cast_fp16)[name = string("x_23_cast_fp16")]; - tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = x_23_cast_fp16)[name = string("inputs_37_cast_fp16")]; - tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; - fp16 var_1591_to_fp16 = const()[name = string("op_1591_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1591_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; - tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50919168)))]; - tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50921280)))]; - fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("input_115_cast_fp16")]; - string var_1611_pad_type_0 = const()[name = string("op_1611_pad_type_0"), val = string("valid")]; - tensor var_1611_strides_0 = const()[name = string("op_1611_strides_0"), val = tensor([1, 1])]; - tensor var_1611_pad_0 = const()[name = string("op_1611_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1611_dilations_0 = const()[name = string("op_1611_dilations_0"), val = tensor([1, 1])]; - int32 var_1611_groups_0 = const()[name = string("op_1611_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50923392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52496320))))[name = string("layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1611_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1611_dilations_0, groups = var_1611_groups_0, pad = var_1611_pad_0, pad_type = var_1611_pad_type_0, strides = var_1611_strides_0, weight = layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = string("op_1611_cast_fp16")]; - string var_1617_pad_type_0 = const()[name = string("op_1617_pad_type_0"), val = string("valid")]; - tensor var_1617_strides_0 = const()[name = string("op_1617_strides_0"), val = tensor([1, 1])]; - tensor var_1617_pad_0 = const()[name = string("op_1617_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1617_dilations_0 = const()[name = string("op_1617_dilations_0"), val = tensor([1, 1])]; - int32 var_1617_groups_0 = const()[name = string("op_1617_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52552448))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52500480))))[name = string("layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1617_cast_fp16 = conv(dilations = var_1617_dilations_0, groups = var_1617_groups_0, pad = var_1617_pad_0, pad_type = var_1617_pad_type_0, strides = var_1617_strides_0, weight = layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_115_cast_fp16)[name = string("op_1617_cast_fp16")]; - tensor input_117_cast_fp16 = add(x = var_1611_cast_fp16, y = var_1617_cast_fp16)[name = string("input_117_cast_fp16")]; - tensor input_119_cast_fp16 = silu(x = input_117_cast_fp16)[name = string("input_119_cast_fp16")]; - string var_1628_pad_type_0 = const()[name = string("op_1628_pad_type_0"), val = string("valid")]; - tensor var_1628_strides_0 = const()[name = string("op_1628_strides_0"), val = tensor([1, 1])]; - tensor var_1628_pad_0 = const()[name = string("op_1628_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1628_dilations_0 = const()[name = string("op_1628_dilations_0"), val = tensor([1, 1])]; - int32 var_1628_groups_0 = const()[name = string("op_1628_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53076800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54649728))))[name = string("layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_1628_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1628_dilations_0, groups = var_1628_groups_0, pad = var_1628_pad_0, pad_type = var_1628_pad_type_0, strides = var_1628_strides_0, weight = layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = string("op_1628_cast_fp16")]; - string var_1634_pad_type_0 = const()[name = string("op_1634_pad_type_0"), val = string("valid")]; - tensor var_1634_strides_0 = const()[name = string("op_1634_strides_0"), val = tensor([1, 1])]; - tensor var_1634_pad_0 = const()[name = string("op_1634_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1634_dilations_0 = const()[name = string("op_1634_dilations_0"), val = tensor([1, 1])]; - int32 var_1634_groups_0 = const()[name = string("op_1634_groups_0"), val = int32(1)]; - tensor layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54723776))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54650816))))[name = string("layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1634_cast_fp16 = conv(dilations = var_1634_dilations_0, groups = var_1634_groups_0, pad = var_1634_pad_0, pad_type = var_1634_pad_type_0, strides = var_1634_strides_0, weight = layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_119_cast_fp16)[name = string("op_1634_cast_fp16")]; - tensor x_25_cast_fp16 = add(x = var_1628_cast_fp16, y = var_1634_cast_fp16)[name = string("x_25_cast_fp16")]; - fp16 var_1636_to_fp16 = const()[name = string("op_1636_to_fp16"), val = fp16(0x1p-1)]; - tensor var_1637_cast_fp16 = mul(x = x_25_cast_fp16, y = var_1636_to_fp16)[name = string("op_1637_cast_fp16")]; - tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = var_1637_cast_fp16)[name = string("inputs_39_cast_fp16")]; - tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; - fp16 var_1647_to_fp16 = const()[name = string("op_1647_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1647_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; - tensor inputs_41_gamma_0_to_fp16 = const()[name = string("inputs_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55248128)))]; - tensor inputs_41_beta_0_to_fp16 = const()[name = string("inputs_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55250240)))]; - fp16 inputs_41_epsilon_0_to_fp16 = const()[name = string("inputs_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_41_cast_fp16 = batch_norm(beta = inputs_41_beta_0_to_fp16, epsilon = inputs_41_epsilon_0_to_fp16, gamma = inputs_41_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("inputs_41_cast_fp16")]; - int32 var_1661 = const()[name = string("op_1661"), val = int32(3)]; - tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; - fp16 var_1692_to_fp16 = const()[name = string("op_1692_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1692_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; - tensor input_121_gamma_0_to_fp16 = const()[name = string("input_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55252352)))]; - tensor input_121_beta_0_to_fp16 = const()[name = string("input_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55254464)))]; - fp16 input_121_epsilon_0_to_fp16 = const()[name = string("input_121_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_121_cast_fp16 = batch_norm(beta = input_121_beta_0_to_fp16, epsilon = input_121_epsilon_0_to_fp16, gamma = input_121_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("input_121_cast_fp16")]; - string var_1712_pad_type_0 = const()[name = string("op_1712_pad_type_0"), val = string("valid")]; - tensor var_1712_strides_0 = const()[name = string("op_1712_strides_0"), val = tensor([1, 1])]; - tensor var_1712_pad_0 = const()[name = string("op_1712_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1712_dilations_0 = const()[name = string("op_1712_dilations_0"), val = tensor([1, 1])]; - int32 var_1712_groups_0 = const()[name = string("op_1712_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55256576))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56829504))))[name = string("layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1712_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1712_dilations_0, groups = var_1712_groups_0, pad = var_1712_pad_0, pad_type = var_1712_pad_type_0, strides = var_1712_strides_0, weight = layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = string("op_1712_cast_fp16")]; - string var_1718_pad_type_0 = const()[name = string("op_1718_pad_type_0"), val = string("valid")]; - tensor var_1718_strides_0 = const()[name = string("op_1718_strides_0"), val = tensor([1, 1])]; - tensor var_1718_pad_0 = const()[name = string("op_1718_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1718_dilations_0 = const()[name = string("op_1718_dilations_0"), val = tensor([1, 1])]; - int32 var_1718_groups_0 = const()[name = string("op_1718_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56878016))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56833664))))[name = string("layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1718_cast_fp16 = conv(dilations = var_1718_dilations_0, groups = var_1718_groups_0, pad = var_1718_pad_0, pad_type = var_1718_pad_type_0, strides = var_1718_strides_0, weight = layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_121_cast_fp16)[name = string("op_1718_cast_fp16")]; - tensor input_123_cast_fp16 = add(x = var_1712_cast_fp16, y = var_1718_cast_fp16)[name = string("input_123_cast_fp16")]; - tensor input_125_cast_fp16 = silu(x = input_123_cast_fp16)[name = string("input_125_cast_fp16")]; - string var_1729_pad_type_0 = const()[name = string("op_1729_pad_type_0"), val = string("valid")]; - tensor var_1729_strides_0 = const()[name = string("op_1729_strides_0"), val = tensor([1, 1])]; - tensor var_1729_pad_0 = const()[name = string("op_1729_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1729_dilations_0 = const()[name = string("op_1729_dilations_0"), val = tensor([1, 1])]; - int32 var_1729_groups_0 = const()[name = string("op_1729_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57402368))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58975296))))[name = string("layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_1729_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1729_dilations_0, groups = var_1729_groups_0, pad = var_1729_pad_0, pad_type = var_1729_pad_type_0, strides = var_1729_strides_0, weight = layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = string("op_1729_cast_fp16")]; - string var_1735_pad_type_0 = const()[name = string("op_1735_pad_type_0"), val = string("valid")]; - tensor var_1735_strides_0 = const()[name = string("op_1735_strides_0"), val = tensor([1, 1])]; - tensor var_1735_pad_0 = const()[name = string("op_1735_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1735_dilations_0 = const()[name = string("op_1735_dilations_0"), val = tensor([1, 1])]; - int32 var_1735_groups_0 = const()[name = string("op_1735_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59057728))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58976384))))[name = string("layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1735_cast_fp16 = conv(dilations = var_1735_dilations_0, groups = var_1735_groups_0, pad = var_1735_pad_0, pad_type = var_1735_pad_type_0, strides = var_1735_strides_0, weight = layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_125_cast_fp16)[name = string("op_1735_cast_fp16")]; - tensor x_27_cast_fp16 = add(x = var_1729_cast_fp16, y = var_1735_cast_fp16)[name = string("x_27_cast_fp16")]; - fp16 var_1737_to_fp16 = const()[name = string("op_1737_to_fp16"), val = fp16(0x1p-1)]; - tensor var_1738_cast_fp16 = mul(x = x_27_cast_fp16, y = var_1737_to_fp16)[name = string("op_1738_cast_fp16")]; - tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = var_1738_cast_fp16)[name = string("inputs_43_cast_fp16")]; - tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; - fp16 var_1748_to_fp16 = const()[name = string("op_1748_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1748_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; - tensor obj_19_gamma_0_to_fp16 = const()[name = string("obj_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59582080)))]; - tensor obj_19_beta_0_to_fp16 = const()[name = string("obj_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59584192)))]; - fp16 obj_19_epsilon_0_to_fp16 = const()[name = string("obj_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_19_cast_fp16 = batch_norm(beta = obj_19_beta_0_to_fp16, epsilon = obj_19_epsilon_0_to_fp16, gamma = obj_19_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("obj_19_cast_fp16")]; - string var_1773_pad_type_0 = const()[name = string("op_1773_pad_type_0"), val = string("valid")]; - tensor var_1773_strides_0 = const()[name = string("op_1773_strides_0"), val = tensor([1, 1])]; - tensor var_1773_pad_0 = const()[name = string("op_1773_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1773_dilations_0 = const()[name = string("op_1773_dilations_0"), val = tensor([1, 1])]; - int32 var_1773_groups_0 = const()[name = string("op_1773_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59586304))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59979584))))[name = string("layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1773_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1773_dilations_0, groups = var_1773_groups_0, pad = var_1773_pad_0, pad_type = var_1773_pad_type_0, strides = var_1773_strides_0, weight = layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = string("op_1773_cast_fp16")]; - string var_1779_pad_type_0 = const()[name = string("op_1779_pad_type_0"), val = string("valid")]; - tensor var_1779_strides_0 = const()[name = string("op_1779_strides_0"), val = tensor([1, 1])]; - tensor var_1779_pad_0 = const()[name = string("op_1779_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1779_dilations_0 = const()[name = string("op_1779_dilations_0"), val = tensor([1, 1])]; - int32 var_1779_groups_0 = const()[name = string("op_1779_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59993600))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59980672))))[name = string("layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1779_cast_fp16 = conv(dilations = var_1779_dilations_0, groups = var_1779_groups_0, pad = var_1779_pad_0, pad_type = var_1779_pad_type_0, strides = var_1779_strides_0, weight = layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = string("op_1779_cast_fp16")]; - tensor query_17_cast_fp16 = add(x = var_1773_cast_fp16, y = var_1779_cast_fp16)[name = string("query_17_cast_fp16")]; - string var_1788_pad_type_0 = const()[name = string("op_1788_pad_type_0"), val = string("valid")]; - tensor var_1788_strides_0 = const()[name = string("op_1788_strides_0"), val = tensor([1, 1])]; - tensor var_1788_pad_0 = const()[name = string("op_1788_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1788_dilations_0 = const()[name = string("op_1788_dilations_0"), val = tensor([1, 1])]; - int32 var_1788_groups_0 = const()[name = string("op_1788_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60124736))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60518016))))[name = string("layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1788_cast_fp16 = conv(dilations = var_1788_dilations_0, groups = var_1788_groups_0, pad = var_1788_pad_0, pad_type = var_1788_pad_type_0, strides = var_1788_strides_0, weight = layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = string("op_1788_cast_fp16")]; - string var_1794_pad_type_0 = const()[name = string("op_1794_pad_type_0"), val = string("valid")]; - tensor var_1794_strides_0 = const()[name = string("op_1794_strides_0"), val = tensor([1, 1])]; - tensor var_1794_pad_0 = const()[name = string("op_1794_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1794_dilations_0 = const()[name = string("op_1794_dilations_0"), val = tensor([1, 1])]; - int32 var_1794_groups_0 = const()[name = string("op_1794_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60530880))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60519104))))[name = string("layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1794_cast_fp16 = conv(dilations = var_1794_dilations_0, groups = var_1794_groups_0, pad = var_1794_pad_0, pad_type = var_1794_pad_type_0, strides = var_1794_strides_0, weight = layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = string("op_1794_cast_fp16")]; - tensor key_9_cast_fp16 = add(x = var_1788_cast_fp16, y = var_1794_cast_fp16)[name = string("key_9_cast_fp16")]; - string var_1804_pad_type_0 = const()[name = string("op_1804_pad_type_0"), val = string("valid")]; - tensor var_1804_strides_0 = const()[name = string("op_1804_strides_0"), val = tensor([1, 1])]; - tensor var_1804_pad_0 = const()[name = string("op_1804_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1804_dilations_0 = const()[name = string("op_1804_dilations_0"), val = tensor([1, 1])]; - int32 var_1804_groups_0 = const()[name = string("op_1804_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60662016))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61055296))))[name = string("layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1804_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1804_dilations_0, groups = var_1804_groups_0, pad = var_1804_pad_0, pad_type = var_1804_pad_type_0, strides = var_1804_strides_0, weight = layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = string("op_1804_cast_fp16")]; - string var_1810_pad_type_0 = const()[name = string("op_1810_pad_type_0"), val = string("valid")]; - tensor var_1810_strides_0 = const()[name = string("op_1810_strides_0"), val = tensor([1, 1])]; - tensor var_1810_pad_0 = const()[name = string("op_1810_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1810_dilations_0 = const()[name = string("op_1810_dilations_0"), val = tensor([1, 1])]; - int32 var_1810_groups_0 = const()[name = string("op_1810_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61065024))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61056384))))[name = string("layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1810_cast_fp16 = conv(dilations = var_1810_dilations_0, groups = var_1810_groups_0, pad = var_1810_pad_0, pad_type = var_1810_pad_type_0, strides = var_1810_strides_0, weight = layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = string("op_1810_cast_fp16")]; - tensor value_9_cast_fp16 = add(x = var_1804_cast_fp16, y = var_1810_cast_fp16)[name = string("value_9_cast_fp16")]; - tensor var_1813_to_fp16 = const()[name = string("op_1813_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61196160)))]; - tensor query_19_cast_fp16 = add(x = query_17_cast_fp16, y = var_1813_to_fp16)[name = string("query_19_cast_fp16")]; - tensor var_1816_to_fp16 = const()[name = string("op_1816_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61198272)))]; - tensor q_with_bias_v_9_cast_fp16 = add(x = query_17_cast_fp16, y = var_1816_to_fp16)[name = string("q_with_bias_v_9_cast_fp16")]; - string var_1826_pad_type_0 = const()[name = string("op_1826_pad_type_0"), val = string("valid")]; - tensor var_1826_strides_0 = const()[name = string("op_1826_strides_0"), val = tensor([1, 1])]; - tensor var_1826_pad_0 = const()[name = string("op_1826_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1826_dilations_0 = const()[name = string("op_1826_dilations_0"), val = tensor([1, 1])]; - int32 var_1826_groups_0 = const()[name = string("op_1826_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61200384))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61593664))))[name = string("layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_1826_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1826_dilations_0, groups = var_1826_groups_0, pad = var_1826_pad_0, pad_type = var_1826_pad_type_0, strides = var_1826_strides_0, weight = layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_1826_cast_fp16")]; - string var_1832_pad_type_0 = const()[name = string("op_1832_pad_type_0"), val = string("valid")]; - tensor var_1832_strides_0 = const()[name = string("op_1832_strides_0"), val = tensor([1, 1])]; - tensor var_1832_pad_0 = const()[name = string("op_1832_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1832_dilations_0 = const()[name = string("op_1832_dilations_0"), val = tensor([1, 1])]; - int32 var_1832_groups_0 = const()[name = string("op_1832_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61629952))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61594752))))[name = string("layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1832_cast_fp16 = conv(dilations = var_1832_dilations_0, groups = var_1832_groups_0, pad = var_1832_pad_0, pad_type = var_1832_pad_type_0, strides = var_1832_strides_0, weight = layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_1832_cast_fp16")]; - tensor p_9_cast_fp16 = add(x = var_1826_cast_fp16, y = var_1832_cast_fp16)[name = string("p_9_cast_fp16")]; - tensor var_1836 = const()[name = string("op_1836"), val = tensor([1, 8, 128, 188])]; - tensor var_1837_cast_fp16 = reshape(shape = var_1836, x = q_with_bias_v_9_cast_fp16)[name = string("op_1837_cast_fp16")]; - tensor var_1838 = const()[name = string("op_1838"), val = tensor([1, 8, 128, -1])]; - tensor var_1839_cast_fp16 = reshape(shape = var_1838, x = p_9_cast_fp16)[name = string("op_1839_cast_fp16")]; - bool matrix_bd_33_transpose_x_0 = const()[name = string("matrix_bd_33_transpose_x_0"), val = bool(true)]; - bool matrix_bd_33_transpose_y_0 = const()[name = string("matrix_bd_33_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_33_cast_fp16 = matmul(transpose_x = matrix_bd_33_transpose_x_0, transpose_y = matrix_bd_33_transpose_y_0, x = var_1837_cast_fp16, y = var_1839_cast_fp16)[name = string("matrix_bd_33_cast_fp16")]; - tensor matrix_bd_35_pad_0 = const()[name = string("matrix_bd_35_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_35_mode_0 = const()[name = string("matrix_bd_35_mode_0"), val = string("constant")]; - fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_35_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = matrix_bd_35_mode_0, pad = matrix_bd_35_pad_0, x = matrix_bd_33_cast_fp16)[name = string("matrix_bd_35_cast_fp16")]; - tensor var_1848 = const()[name = string("op_1848"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_37_cast_fp16 = reshape(shape = var_1848, x = matrix_bd_35_cast_fp16)[name = string("matrix_bd_37_cast_fp16")]; - tensor var_1852_begin_0 = const()[name = string("op_1852_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_1852_end_0 = const()[name = string("op_1852_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_1852_end_mask_0 = const()[name = string("op_1852_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_1852_cast_fp16 = slice_by_index(begin = var_1852_begin_0, end = var_1852_end_0, end_mask = var_1852_end_mask_0, x = matrix_bd_37_cast_fp16)[name = string("op_1852_cast_fp16")]; - tensor var_1853 = const()[name = string("op_1853"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_39_cast_fp16 = reshape(shape = var_1853, x = var_1852_cast_fp16)[name = string("matrix_bd_39_cast_fp16")]; - tensor var_1858_begin_0 = const()[name = string("op_1858_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1858_end_0 = const()[name = string("op_1858_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_1858_end_mask_0 = const()[name = string("op_1858_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_1858_cast_fp16 = slice_by_index(begin = var_1858_begin_0, end = var_1858_end_0, end_mask = var_1858_end_mask_0, x = matrix_bd_39_cast_fp16)[name = string("op_1858_cast_fp16")]; - fp16 var_1859_to_fp16 = const()[name = string("op_1859_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_9_cast_fp16 = mul(x = var_1858_cast_fp16, y = var_1859_to_fp16)[name = string("qk_mask_9_cast_fp16")]; - tensor var_1863 = const()[name = string("op_1863"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_9_cast_fp16 = reshape(shape = var_1863, x = query_19_cast_fp16)[name = string("mh_q_9_cast_fp16")]; - fp16 var_1865_to_fp16 = const()[name = string("op_1865_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_1866_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_1865_to_fp16)[name = string("op_1866_cast_fp16")]; - tensor var_1869 = const()[name = string("op_1869"), val = tensor([1, 8, 128, 188])]; - tensor var_1870_cast_fp16 = reshape(shape = var_1869, x = key_9_cast_fp16)[name = string("op_1870_cast_fp16")]; - bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; - bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; - tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_1866_cast_fp16, y = var_1870_cast_fp16)[name = string("mh_w_17_cast_fp16")]; - tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = qk_mask_9_cast_fp16)[name = string("mh_w_19_cast_fp16")]; - tensor var_1874_cast_fp16 = softmax(axis = var_1661, x = mh_w_19_cast_fp16)[name = string("op_1874_cast_fp16")]; - tensor var_1875 = const()[name = string("op_1875"), val = tensor([1, 8, 128, 188])]; - tensor var_1876_cast_fp16 = reshape(shape = var_1875, x = value_9_cast_fp16)[name = string("op_1876_cast_fp16")]; - bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; - bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; - tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_1876_cast_fp16, y = var_1874_cast_fp16)[name = string("attn_9_cast_fp16")]; - tensor var_1879 = const()[name = string("op_1879"), val = tensor([1, 1024, 1, 188])]; - tensor input_127_cast_fp16 = reshape(shape = var_1879, x = attn_9_cast_fp16)[name = string("input_127_cast_fp16")]; - string var_1889_pad_type_0 = const()[name = string("op_1889_pad_type_0"), val = string("valid")]; - tensor var_1889_strides_0 = const()[name = string("op_1889_strides_0"), val = tensor([1, 1])]; - tensor var_1889_pad_0 = const()[name = string("op_1889_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1889_dilations_0 = const()[name = string("op_1889_dilations_0"), val = tensor([1, 1])]; - int32 var_1889_groups_0 = const()[name = string("op_1889_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61761088))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62154368))))[name = string("layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_1889_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1889_dilations_0, groups = var_1889_groups_0, pad = var_1889_pad_0, pad_type = var_1889_pad_type_0, strides = var_1889_strides_0, weight = layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = string("op_1889_cast_fp16")]; - string var_1895_pad_type_0 = const()[name = string("op_1895_pad_type_0"), val = string("valid")]; - tensor var_1895_strides_0 = const()[name = string("op_1895_strides_0"), val = tensor([1, 1])]; - tensor var_1895_pad_0 = const()[name = string("op_1895_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1895_dilations_0 = const()[name = string("op_1895_dilations_0"), val = tensor([1, 1])]; - int32 var_1895_groups_0 = const()[name = string("op_1895_groups_0"), val = int32(1)]; - tensor layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62164160))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62155456))))[name = string("layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1895_cast_fp16 = conv(dilations = var_1895_dilations_0, groups = var_1895_groups_0, pad = var_1895_pad_0, pad_type = var_1895_pad_type_0, strides = var_1895_strides_0, weight = layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_127_cast_fp16)[name = string("op_1895_cast_fp16")]; - tensor obj_21_cast_fp16 = add(x = var_1889_cast_fp16, y = var_1895_cast_fp16)[name = string("obj_21_cast_fp16")]; - tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_45_cast_fp16")]; - tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; - fp16 var_1906_to_fp16 = const()[name = string("op_1906_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1906_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; - tensor input_129_gamma_0_to_fp16 = const()[name = string("input_129_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62295296)))]; - tensor input_129_beta_0_to_fp16 = const()[name = string("input_129_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62297408)))]; - fp16 input_129_epsilon_0_to_fp16 = const()[name = string("input_129_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_129_cast_fp16 = batch_norm(beta = input_129_beta_0_to_fp16, epsilon = input_129_epsilon_0_to_fp16, gamma = input_129_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("input_129_cast_fp16")]; - string var_1927_pad_type_0 = const()[name = string("op_1927_pad_type_0"), val = string("valid")]; - tensor var_1927_strides_0 = const()[name = string("op_1927_strides_0"), val = tensor([1, 1])]; - tensor var_1927_pad_0 = const()[name = string("op_1927_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1927_dilations_0 = const()[name = string("op_1927_dilations_0"), val = tensor([1, 1])]; - int32 var_1927_groups_0 = const()[name = string("op_1927_groups_0"), val = int32(1)]; - tensor layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62299520))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63086016))))[name = string("layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1927_cast_fp16 = conv(dilations = var_1927_dilations_0, groups = var_1927_groups_0, pad = var_1927_pad_0, pad_type = var_1927_pad_type_0, strides = var_1927_strides_0, weight = layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = string("op_1927_cast_fp16")]; - string var_1933_pad_type_0 = const()[name = string("op_1933_pad_type_0"), val = string("valid")]; - tensor var_1933_strides_0 = const()[name = string("op_1933_strides_0"), val = tensor([1, 1])]; - tensor var_1933_pad_0 = const()[name = string("op_1933_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1933_dilations_0 = const()[name = string("op_1933_dilations_0"), val = tensor([1, 1])]; - int32 var_1933_groups_0 = const()[name = string("op_1933_groups_0"), val = int32(1)]; - tensor layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63109888))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63088128))))[name = string("layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1933_cast_fp16 = conv(dilations = var_1933_dilations_0, groups = var_1933_groups_0, pad = var_1933_pad_0, pad_type = var_1933_pad_type_0, strides = var_1933_strides_0, weight = layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_129_cast_fp16)[name = string("op_1933_cast_fp16")]; - tensor input_131_cast_fp16 = add(x = var_1927_cast_fp16, y = var_1933_cast_fp16)[name = string("input_131_cast_fp16")]; - int32 input_133_split_num_splits_0 = const()[name = string("input_133_split_num_splits_0"), val = int32(2)]; - int32 input_133_split_axis_0 = const()[name = string("input_133_split_axis_0"), val = int32(1)]; - tensor input_133_split_cast_fp16_0, tensor input_133_split_cast_fp16_1 = split(axis = input_133_split_axis_0, num_splits = input_133_split_num_splits_0, x = input_131_cast_fp16)[name = string("input_133_split_cast_fp16")]; - tensor input_133_split_1_sigmoid_cast_fp16 = sigmoid(x = input_133_split_cast_fp16_1)[name = string("input_133_split_1_sigmoid_cast_fp16")]; - tensor input_133_cast_fp16 = mul(x = input_133_split_cast_fp16_0, y = input_133_split_1_sigmoid_cast_fp16)[name = string("input_133_cast_fp16")]; - string input_135_pad_type_0 = const()[name = string("input_135_pad_type_0"), val = string("custom")]; - tensor input_135_pad_0 = const()[name = string("input_135_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_135_groups_0 = const()[name = string("input_135_groups_0"), val = int32(1024)]; - tensor input_135_strides_0 = const()[name = string("input_135_strides_0"), val = tensor([1, 1])]; - tensor input_135_dilations_0 = const()[name = string("input_135_dilations_0"), val = tensor([1, 1])]; - tensor const_276_to_fp16 = const()[name = string("const_276_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63372096)))]; - tensor const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63390592)))]; - tensor input_137_cast_fp16 = conv(bias = const_277_to_fp16, dilations = input_135_dilations_0, groups = input_135_groups_0, pad = input_135_pad_0, pad_type = input_135_pad_type_0, strides = input_135_strides_0, weight = const_276_to_fp16, x = input_133_cast_fp16)[name = string("input_137_cast_fp16")]; - tensor input_139_cast_fp16 = silu(x = input_137_cast_fp16)[name = string("input_139_cast_fp16")]; - string var_1955_pad_type_0 = const()[name = string("op_1955_pad_type_0"), val = string("valid")]; - tensor var_1955_strides_0 = const()[name = string("op_1955_strides_0"), val = tensor([1, 1])]; - tensor var_1955_pad_0 = const()[name = string("op_1955_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1955_dilations_0 = const()[name = string("op_1955_dilations_0"), val = tensor([1, 1])]; - int32 var_1955_groups_0 = const()[name = string("op_1955_groups_0"), val = int32(1)]; - tensor layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63392704))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63785984))))[name = string("layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_1955_cast_fp16 = conv(dilations = var_1955_dilations_0, groups = var_1955_groups_0, pad = var_1955_pad_0, pad_type = var_1955_pad_type_0, strides = var_1955_strides_0, weight = layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = string("op_1955_cast_fp16")]; - string var_1961_pad_type_0 = const()[name = string("op_1961_pad_type_0"), val = string("valid")]; - tensor var_1961_strides_0 = const()[name = string("op_1961_strides_0"), val = tensor([1, 1])]; - tensor var_1961_pad_0 = const()[name = string("op_1961_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1961_dilations_0 = const()[name = string("op_1961_dilations_0"), val = tensor([1, 1])]; - int32 var_1961_groups_0 = const()[name = string("op_1961_groups_0"), val = int32(1)]; - tensor layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63796800))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63787072))))[name = string("layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1961_cast_fp16 = conv(dilations = var_1961_dilations_0, groups = var_1961_groups_0, pad = var_1961_pad_0, pad_type = var_1961_pad_type_0, strides = var_1961_strides_0, weight = layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_139_cast_fp16)[name = string("op_1961_cast_fp16")]; - tensor x_29_cast_fp16 = add(x = var_1955_cast_fp16, y = var_1961_cast_fp16)[name = string("x_29_cast_fp16")]; - tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = x_29_cast_fp16)[name = string("inputs_47_cast_fp16")]; - tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; - fp16 var_1972_to_fp16 = const()[name = string("op_1972_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1972_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; - tensor input_141_gamma_0_to_fp16 = const()[name = string("input_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63927936)))]; - tensor input_141_beta_0_to_fp16 = const()[name = string("input_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63930048)))]; - fp16 input_141_epsilon_0_to_fp16 = const()[name = string("input_141_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_141_cast_fp16 = batch_norm(beta = input_141_beta_0_to_fp16, epsilon = input_141_epsilon_0_to_fp16, gamma = input_141_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_141_cast_fp16")]; - string var_1992_pad_type_0 = const()[name = string("op_1992_pad_type_0"), val = string("valid")]; - tensor var_1992_strides_0 = const()[name = string("op_1992_strides_0"), val = tensor([1, 1])]; - tensor var_1992_pad_0 = const()[name = string("op_1992_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1992_dilations_0 = const()[name = string("op_1992_dilations_0"), val = tensor([1, 1])]; - int32 var_1992_groups_0 = const()[name = string("op_1992_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63932160))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65505088))))[name = string("layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_1992_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1992_dilations_0, groups = var_1992_groups_0, pad = var_1992_pad_0, pad_type = var_1992_pad_type_0, strides = var_1992_strides_0, weight = layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = string("op_1992_cast_fp16")]; - string var_1998_pad_type_0 = const()[name = string("op_1998_pad_type_0"), val = string("valid")]; - tensor var_1998_strides_0 = const()[name = string("op_1998_strides_0"), val = tensor([1, 1])]; - tensor var_1998_pad_0 = const()[name = string("op_1998_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_1998_dilations_0 = const()[name = string("op_1998_dilations_0"), val = tensor([1, 1])]; - int32 var_1998_groups_0 = const()[name = string("op_1998_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65560896))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65509248))))[name = string("layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_1998_cast_fp16 = conv(dilations = var_1998_dilations_0, groups = var_1998_groups_0, pad = var_1998_pad_0, pad_type = var_1998_pad_type_0, strides = var_1998_strides_0, weight = layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_141_cast_fp16)[name = string("op_1998_cast_fp16")]; - tensor input_143_cast_fp16 = add(x = var_1992_cast_fp16, y = var_1998_cast_fp16)[name = string("input_143_cast_fp16")]; - tensor input_145_cast_fp16 = silu(x = input_143_cast_fp16)[name = string("input_145_cast_fp16")]; - string var_2009_pad_type_0 = const()[name = string("op_2009_pad_type_0"), val = string("valid")]; - tensor var_2009_strides_0 = const()[name = string("op_2009_strides_0"), val = tensor([1, 1])]; - tensor var_2009_pad_0 = const()[name = string("op_2009_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2009_dilations_0 = const()[name = string("op_2009_dilations_0"), val = tensor([1, 1])]; - int32 var_2009_groups_0 = const()[name = string("op_2009_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66085248))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67658176))))[name = string("layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2009_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2009_dilations_0, groups = var_2009_groups_0, pad = var_2009_pad_0, pad_type = var_2009_pad_type_0, strides = var_2009_strides_0, weight = layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = string("op_2009_cast_fp16")]; - string var_2015_pad_type_0 = const()[name = string("op_2015_pad_type_0"), val = string("valid")]; - tensor var_2015_strides_0 = const()[name = string("op_2015_strides_0"), val = tensor([1, 1])]; - tensor var_2015_pad_0 = const()[name = string("op_2015_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2015_dilations_0 = const()[name = string("op_2015_dilations_0"), val = tensor([1, 1])]; - int32 var_2015_groups_0 = const()[name = string("op_2015_groups_0"), val = int32(1)]; - tensor layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67733888))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67659264))))[name = string("layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2015_cast_fp16 = conv(dilations = var_2015_dilations_0, groups = var_2015_groups_0, pad = var_2015_pad_0, pad_type = var_2015_pad_type_0, strides = var_2015_strides_0, weight = layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_145_cast_fp16)[name = string("op_2015_cast_fp16")]; - tensor x_31_cast_fp16 = add(x = var_2009_cast_fp16, y = var_2015_cast_fp16)[name = string("x_31_cast_fp16")]; - fp16 var_2017_to_fp16 = const()[name = string("op_2017_to_fp16"), val = fp16(0x1p-1)]; - tensor var_2018_cast_fp16 = mul(x = x_31_cast_fp16, y = var_2017_to_fp16)[name = string("op_2018_cast_fp16")]; - tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = var_2018_cast_fp16)[name = string("inputs_49_cast_fp16")]; - tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; - fp16 var_2028_to_fp16 = const()[name = string("op_2028_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_2028_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; - tensor inputs_51_gamma_0_to_fp16 = const()[name = string("inputs_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68258240)))]; - tensor inputs_51_beta_0_to_fp16 = const()[name = string("inputs_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68260352)))]; - fp16 inputs_51_epsilon_0_to_fp16 = const()[name = string("inputs_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_51_cast_fp16 = batch_norm(beta = inputs_51_beta_0_to_fp16, epsilon = inputs_51_epsilon_0_to_fp16, gamma = inputs_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("inputs_51_cast_fp16")]; - int32 var_2042 = const()[name = string("op_2042"), val = int32(3)]; - tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; - fp16 var_2073_to_fp16 = const()[name = string("op_2073_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_2073_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; - tensor input_147_gamma_0_to_fp16 = const()[name = string("input_147_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68262464)))]; - tensor input_147_beta_0_to_fp16 = const()[name = string("input_147_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68264576)))]; - fp16 input_147_epsilon_0_to_fp16 = const()[name = string("input_147_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_147_cast_fp16 = batch_norm(beta = input_147_beta_0_to_fp16, epsilon = input_147_epsilon_0_to_fp16, gamma = input_147_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("input_147_cast_fp16")]; - string var_2093_pad_type_0 = const()[name = string("op_2093_pad_type_0"), val = string("valid")]; - tensor var_2093_strides_0 = const()[name = string("op_2093_strides_0"), val = tensor([1, 1])]; - tensor var_2093_pad_0 = const()[name = string("op_2093_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2093_dilations_0 = const()[name = string("op_2093_dilations_0"), val = tensor([1, 1])]; - int32 var_2093_groups_0 = const()[name = string("op_2093_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68266688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69839616))))[name = string("layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_2093_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2093_dilations_0, groups = var_2093_groups_0, pad = var_2093_pad_0, pad_type = var_2093_pad_type_0, strides = var_2093_strides_0, weight = layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = string("op_2093_cast_fp16")]; - string var_2099_pad_type_0 = const()[name = string("op_2099_pad_type_0"), val = string("valid")]; - tensor var_2099_strides_0 = const()[name = string("op_2099_strides_0"), val = tensor([1, 1])]; - tensor var_2099_pad_0 = const()[name = string("op_2099_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2099_dilations_0 = const()[name = string("op_2099_dilations_0"), val = tensor([1, 1])]; - int32 var_2099_groups_0 = const()[name = string("op_2099_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69884224))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69843776))))[name = string("layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2099_cast_fp16 = conv(dilations = var_2099_dilations_0, groups = var_2099_groups_0, pad = var_2099_pad_0, pad_type = var_2099_pad_type_0, strides = var_2099_strides_0, weight = layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_147_cast_fp16)[name = string("op_2099_cast_fp16")]; - tensor input_149_cast_fp16 = add(x = var_2093_cast_fp16, y = var_2099_cast_fp16)[name = string("input_149_cast_fp16")]; - tensor input_151_cast_fp16 = silu(x = input_149_cast_fp16)[name = string("input_151_cast_fp16")]; - string var_2110_pad_type_0 = const()[name = string("op_2110_pad_type_0"), val = string("valid")]; - tensor var_2110_strides_0 = const()[name = string("op_2110_strides_0"), val = tensor([1, 1])]; - tensor var_2110_pad_0 = const()[name = string("op_2110_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2110_dilations_0 = const()[name = string("op_2110_dilations_0"), val = tensor([1, 1])]; - int32 var_2110_groups_0 = const()[name = string("op_2110_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70408576))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71981504))))[name = string("layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2110_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2110_dilations_0, groups = var_2110_groups_0, pad = var_2110_pad_0, pad_type = var_2110_pad_type_0, strides = var_2110_strides_0, weight = layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = string("op_2110_cast_fp16")]; - string var_2116_pad_type_0 = const()[name = string("op_2116_pad_type_0"), val = string("valid")]; - tensor var_2116_strides_0 = const()[name = string("op_2116_strides_0"), val = tensor([1, 1])]; - tensor var_2116_pad_0 = const()[name = string("op_2116_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2116_dilations_0 = const()[name = string("op_2116_dilations_0"), val = tensor([1, 1])]; - int32 var_2116_groups_0 = const()[name = string("op_2116_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72062464))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71982592))))[name = string("layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2116_cast_fp16 = conv(dilations = var_2116_dilations_0, groups = var_2116_groups_0, pad = var_2116_pad_0, pad_type = var_2116_pad_type_0, strides = var_2116_strides_0, weight = layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_151_cast_fp16)[name = string("op_2116_cast_fp16")]; - tensor x_33_cast_fp16 = add(x = var_2110_cast_fp16, y = var_2116_cast_fp16)[name = string("x_33_cast_fp16")]; - fp16 var_2118_to_fp16 = const()[name = string("op_2118_to_fp16"), val = fp16(0x1p-1)]; - tensor var_2119_cast_fp16 = mul(x = x_33_cast_fp16, y = var_2118_to_fp16)[name = string("op_2119_cast_fp16")]; - tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = var_2119_cast_fp16)[name = string("inputs_53_cast_fp16")]; - tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; - fp16 var_2129_to_fp16 = const()[name = string("op_2129_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_2129_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; - tensor obj_23_gamma_0_to_fp16 = const()[name = string("obj_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72586816)))]; - tensor obj_23_beta_0_to_fp16 = const()[name = string("obj_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72588928)))]; - fp16 obj_23_epsilon_0_to_fp16 = const()[name = string("obj_23_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("obj_23_cast_fp16")]; - string var_2154_pad_type_0 = const()[name = string("op_2154_pad_type_0"), val = string("valid")]; - tensor var_2154_strides_0 = const()[name = string("op_2154_strides_0"), val = tensor([1, 1])]; - tensor var_2154_pad_0 = const()[name = string("op_2154_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2154_dilations_0 = const()[name = string("op_2154_dilations_0"), val = tensor([1, 1])]; - int32 var_2154_groups_0 = const()[name = string("op_2154_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72591040))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72984320))))[name = string("layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2154_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2154_dilations_0, groups = var_2154_groups_0, pad = var_2154_pad_0, pad_type = var_2154_pad_type_0, strides = var_2154_strides_0, weight = layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = string("op_2154_cast_fp16")]; - string var_2160_pad_type_0 = const()[name = string("op_2160_pad_type_0"), val = string("valid")]; - tensor var_2160_strides_0 = const()[name = string("op_2160_strides_0"), val = tensor([1, 1])]; - tensor var_2160_pad_0 = const()[name = string("op_2160_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2160_dilations_0 = const()[name = string("op_2160_dilations_0"), val = tensor([1, 1])]; - int32 var_2160_groups_0 = const()[name = string("op_2160_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72996992))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72985408))))[name = string("layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2160_cast_fp16 = conv(dilations = var_2160_dilations_0, groups = var_2160_groups_0, pad = var_2160_pad_0, pad_type = var_2160_pad_type_0, strides = var_2160_strides_0, weight = layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = string("op_2160_cast_fp16")]; - tensor query_21_cast_fp16 = add(x = var_2154_cast_fp16, y = var_2160_cast_fp16)[name = string("query_21_cast_fp16")]; - string var_2169_pad_type_0 = const()[name = string("op_2169_pad_type_0"), val = string("valid")]; - tensor var_2169_strides_0 = const()[name = string("op_2169_strides_0"), val = tensor([1, 1])]; - tensor var_2169_pad_0 = const()[name = string("op_2169_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2169_dilations_0 = const()[name = string("op_2169_dilations_0"), val = tensor([1, 1])]; - int32 var_2169_groups_0 = const()[name = string("op_2169_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73128128))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73521408))))[name = string("layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2169_cast_fp16 = conv(dilations = var_2169_dilations_0, groups = var_2169_groups_0, pad = var_2169_pad_0, pad_type = var_2169_pad_type_0, strides = var_2169_strides_0, weight = layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = string("op_2169_cast_fp16")]; - string var_2175_pad_type_0 = const()[name = string("op_2175_pad_type_0"), val = string("valid")]; - tensor var_2175_strides_0 = const()[name = string("op_2175_strides_0"), val = tensor([1, 1])]; - tensor var_2175_pad_0 = const()[name = string("op_2175_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2175_dilations_0 = const()[name = string("op_2175_dilations_0"), val = tensor([1, 1])]; - int32 var_2175_groups_0 = const()[name = string("op_2175_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73536448))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73522496))))[name = string("layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2175_cast_fp16 = conv(dilations = var_2175_dilations_0, groups = var_2175_groups_0, pad = var_2175_pad_0, pad_type = var_2175_pad_type_0, strides = var_2175_strides_0, weight = layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = string("op_2175_cast_fp16")]; - tensor key_11_cast_fp16 = add(x = var_2169_cast_fp16, y = var_2175_cast_fp16)[name = string("key_11_cast_fp16")]; - string var_2185_pad_type_0 = const()[name = string("op_2185_pad_type_0"), val = string("valid")]; - tensor var_2185_strides_0 = const()[name = string("op_2185_strides_0"), val = tensor([1, 1])]; - tensor var_2185_pad_0 = const()[name = string("op_2185_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2185_dilations_0 = const()[name = string("op_2185_dilations_0"), val = tensor([1, 1])]; - int32 var_2185_groups_0 = const()[name = string("op_2185_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73667584))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74060864))))[name = string("layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2185_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2185_dilations_0, groups = var_2185_groups_0, pad = var_2185_pad_0, pad_type = var_2185_pad_type_0, strides = var_2185_strides_0, weight = layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = string("op_2185_cast_fp16")]; - string var_2191_pad_type_0 = const()[name = string("op_2191_pad_type_0"), val = string("valid")]; - tensor var_2191_strides_0 = const()[name = string("op_2191_strides_0"), val = tensor([1, 1])]; - tensor var_2191_pad_0 = const()[name = string("op_2191_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2191_dilations_0 = const()[name = string("op_2191_dilations_0"), val = tensor([1, 1])]; - int32 var_2191_groups_0 = const()[name = string("op_2191_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74070848))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74061952))))[name = string("layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2191_cast_fp16 = conv(dilations = var_2191_dilations_0, groups = var_2191_groups_0, pad = var_2191_pad_0, pad_type = var_2191_pad_type_0, strides = var_2191_strides_0, weight = layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = string("op_2191_cast_fp16")]; - tensor value_11_cast_fp16 = add(x = var_2185_cast_fp16, y = var_2191_cast_fp16)[name = string("value_11_cast_fp16")]; - tensor var_2194_to_fp16 = const()[name = string("op_2194_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74201984)))]; - tensor query_23_cast_fp16 = add(x = query_21_cast_fp16, y = var_2194_to_fp16)[name = string("query_23_cast_fp16")]; - tensor var_2197_to_fp16 = const()[name = string("op_2197_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74204096)))]; - tensor q_with_bias_v_11_cast_fp16 = add(x = query_21_cast_fp16, y = var_2197_to_fp16)[name = string("q_with_bias_v_11_cast_fp16")]; - string var_2207_pad_type_0 = const()[name = string("op_2207_pad_type_0"), val = string("valid")]; - tensor var_2207_strides_0 = const()[name = string("op_2207_strides_0"), val = tensor([1, 1])]; - tensor var_2207_pad_0 = const()[name = string("op_2207_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2207_dilations_0 = const()[name = string("op_2207_dilations_0"), val = tensor([1, 1])]; - int32 var_2207_groups_0 = const()[name = string("op_2207_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74206208))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74599488))))[name = string("layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_2207_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2207_dilations_0, groups = var_2207_groups_0, pad = var_2207_pad_0, pad_type = var_2207_pad_type_0, strides = var_2207_strides_0, weight = layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_2207_cast_fp16")]; - string var_2213_pad_type_0 = const()[name = string("op_2213_pad_type_0"), val = string("valid")]; - tensor var_2213_strides_0 = const()[name = string("op_2213_strides_0"), val = tensor([1, 1])]; - tensor var_2213_pad_0 = const()[name = string("op_2213_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2213_dilations_0 = const()[name = string("op_2213_dilations_0"), val = tensor([1, 1])]; - int32 var_2213_groups_0 = const()[name = string("op_2213_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74635264))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74600576))))[name = string("layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2213_cast_fp16 = conv(dilations = var_2213_dilations_0, groups = var_2213_groups_0, pad = var_2213_pad_0, pad_type = var_2213_pad_type_0, strides = var_2213_strides_0, weight = layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_2213_cast_fp16")]; - tensor p_11_cast_fp16 = add(x = var_2207_cast_fp16, y = var_2213_cast_fp16)[name = string("p_11_cast_fp16")]; - tensor var_2217 = const()[name = string("op_2217"), val = tensor([1, 8, 128, 188])]; - tensor var_2218_cast_fp16 = reshape(shape = var_2217, x = q_with_bias_v_11_cast_fp16)[name = string("op_2218_cast_fp16")]; - tensor var_2219 = const()[name = string("op_2219"), val = tensor([1, 8, 128, -1])]; - tensor var_2220_cast_fp16 = reshape(shape = var_2219, x = p_11_cast_fp16)[name = string("op_2220_cast_fp16")]; - bool matrix_bd_41_transpose_x_0 = const()[name = string("matrix_bd_41_transpose_x_0"), val = bool(true)]; - bool matrix_bd_41_transpose_y_0 = const()[name = string("matrix_bd_41_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_41_cast_fp16 = matmul(transpose_x = matrix_bd_41_transpose_x_0, transpose_y = matrix_bd_41_transpose_y_0, x = var_2218_cast_fp16, y = var_2220_cast_fp16)[name = string("matrix_bd_41_cast_fp16")]; - tensor matrix_bd_43_pad_0 = const()[name = string("matrix_bd_43_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_43_mode_0 = const()[name = string("matrix_bd_43_mode_0"), val = string("constant")]; - fp16 const_65_to_fp16 = const()[name = string("const_65_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_43_cast_fp16 = pad(constant_val = const_65_to_fp16, mode = matrix_bd_43_mode_0, pad = matrix_bd_43_pad_0, x = matrix_bd_41_cast_fp16)[name = string("matrix_bd_43_cast_fp16")]; - tensor var_2229 = const()[name = string("op_2229"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_45_cast_fp16 = reshape(shape = var_2229, x = matrix_bd_43_cast_fp16)[name = string("matrix_bd_45_cast_fp16")]; - tensor var_2233_begin_0 = const()[name = string("op_2233_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_2233_end_0 = const()[name = string("op_2233_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_2233_end_mask_0 = const()[name = string("op_2233_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_2233_cast_fp16 = slice_by_index(begin = var_2233_begin_0, end = var_2233_end_0, end_mask = var_2233_end_mask_0, x = matrix_bd_45_cast_fp16)[name = string("op_2233_cast_fp16")]; - tensor var_2234 = const()[name = string("op_2234"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_47_cast_fp16 = reshape(shape = var_2234, x = var_2233_cast_fp16)[name = string("matrix_bd_47_cast_fp16")]; - tensor var_2239_begin_0 = const()[name = string("op_2239_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2239_end_0 = const()[name = string("op_2239_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_2239_end_mask_0 = const()[name = string("op_2239_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_2239_cast_fp16 = slice_by_index(begin = var_2239_begin_0, end = var_2239_end_0, end_mask = var_2239_end_mask_0, x = matrix_bd_47_cast_fp16)[name = string("op_2239_cast_fp16")]; - fp16 var_2240_to_fp16 = const()[name = string("op_2240_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_11_cast_fp16 = mul(x = var_2239_cast_fp16, y = var_2240_to_fp16)[name = string("qk_mask_11_cast_fp16")]; - tensor var_2244 = const()[name = string("op_2244"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_11_cast_fp16 = reshape(shape = var_2244, x = query_23_cast_fp16)[name = string("mh_q_11_cast_fp16")]; - fp16 var_2246_to_fp16 = const()[name = string("op_2246_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_2247_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_2246_to_fp16)[name = string("op_2247_cast_fp16")]; - tensor var_2250 = const()[name = string("op_2250"), val = tensor([1, 8, 128, 188])]; - tensor var_2251_cast_fp16 = reshape(shape = var_2250, x = key_11_cast_fp16)[name = string("op_2251_cast_fp16")]; - bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; - bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; - tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_2247_cast_fp16, y = var_2251_cast_fp16)[name = string("mh_w_21_cast_fp16")]; - tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = qk_mask_11_cast_fp16)[name = string("mh_w_23_cast_fp16")]; - tensor var_2255_cast_fp16 = softmax(axis = var_2042, x = mh_w_23_cast_fp16)[name = string("op_2255_cast_fp16")]; - tensor var_2256 = const()[name = string("op_2256"), val = tensor([1, 8, 128, 188])]; - tensor var_2257_cast_fp16 = reshape(shape = var_2256, x = value_11_cast_fp16)[name = string("op_2257_cast_fp16")]; - bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; - bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; - tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_2257_cast_fp16, y = var_2255_cast_fp16)[name = string("attn_11_cast_fp16")]; - tensor var_2260 = const()[name = string("op_2260"), val = tensor([1, 1024, 1, 188])]; - tensor input_153_cast_fp16 = reshape(shape = var_2260, x = attn_11_cast_fp16)[name = string("input_153_cast_fp16")]; - string var_2270_pad_type_0 = const()[name = string("op_2270_pad_type_0"), val = string("valid")]; - tensor var_2270_strides_0 = const()[name = string("op_2270_strides_0"), val = tensor([1, 1])]; - tensor var_2270_pad_0 = const()[name = string("op_2270_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2270_dilations_0 = const()[name = string("op_2270_dilations_0"), val = tensor([1, 1])]; - int32 var_2270_groups_0 = const()[name = string("op_2270_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74766400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75159680))))[name = string("layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2270_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2270_dilations_0, groups = var_2270_groups_0, pad = var_2270_pad_0, pad_type = var_2270_pad_type_0, strides = var_2270_strides_0, weight = layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = string("op_2270_cast_fp16")]; - string var_2276_pad_type_0 = const()[name = string("op_2276_pad_type_0"), val = string("valid")]; - tensor var_2276_strides_0 = const()[name = string("op_2276_strides_0"), val = tensor([1, 1])]; - tensor var_2276_pad_0 = const()[name = string("op_2276_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2276_dilations_0 = const()[name = string("op_2276_dilations_0"), val = tensor([1, 1])]; - int32 var_2276_groups_0 = const()[name = string("op_2276_groups_0"), val = int32(1)]; - tensor layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75169408))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75160768))))[name = string("layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2276_cast_fp16 = conv(dilations = var_2276_dilations_0, groups = var_2276_groups_0, pad = var_2276_pad_0, pad_type = var_2276_pad_type_0, strides = var_2276_strides_0, weight = layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_153_cast_fp16)[name = string("op_2276_cast_fp16")]; - tensor obj_25_cast_fp16 = add(x = var_2270_cast_fp16, y = var_2276_cast_fp16)[name = string("obj_25_cast_fp16")]; - tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_25_cast_fp16)[name = string("inputs_55_cast_fp16")]; - tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; - fp16 var_2287_to_fp16 = const()[name = string("op_2287_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2287_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; - tensor input_155_gamma_0_to_fp16 = const()[name = string("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75300544)))]; - tensor input_155_beta_0_to_fp16 = const()[name = string("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75302656)))]; - fp16 input_155_epsilon_0_to_fp16 = const()[name = string("input_155_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("input_155_cast_fp16")]; - string var_2308_pad_type_0 = const()[name = string("op_2308_pad_type_0"), val = string("valid")]; - tensor var_2308_strides_0 = const()[name = string("op_2308_strides_0"), val = tensor([1, 1])]; - tensor var_2308_pad_0 = const()[name = string("op_2308_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2308_dilations_0 = const()[name = string("op_2308_dilations_0"), val = tensor([1, 1])]; - int32 var_2308_groups_0 = const()[name = string("op_2308_groups_0"), val = int32(1)]; - tensor layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75304768))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76091264))))[name = string("layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_2308_cast_fp16 = conv(dilations = var_2308_dilations_0, groups = var_2308_groups_0, pad = var_2308_pad_0, pad_type = var_2308_pad_type_0, strides = var_2308_strides_0, weight = layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = string("op_2308_cast_fp16")]; - string var_2314_pad_type_0 = const()[name = string("op_2314_pad_type_0"), val = string("valid")]; - tensor var_2314_strides_0 = const()[name = string("op_2314_strides_0"), val = tensor([1, 1])]; - tensor var_2314_pad_0 = const()[name = string("op_2314_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2314_dilations_0 = const()[name = string("op_2314_dilations_0"), val = tensor([1, 1])]; - int32 var_2314_groups_0 = const()[name = string("op_2314_groups_0"), val = int32(1)]; - tensor layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76113984))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76093376))))[name = string("layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2314_cast_fp16 = conv(dilations = var_2314_dilations_0, groups = var_2314_groups_0, pad = var_2314_pad_0, pad_type = var_2314_pad_type_0, strides = var_2314_strides_0, weight = layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_155_cast_fp16)[name = string("op_2314_cast_fp16")]; - tensor input_157_cast_fp16 = add(x = var_2308_cast_fp16, y = var_2314_cast_fp16)[name = string("input_157_cast_fp16")]; - int32 input_159_split_num_splits_0 = const()[name = string("input_159_split_num_splits_0"), val = int32(2)]; - int32 input_159_split_axis_0 = const()[name = string("input_159_split_axis_0"), val = int32(1)]; - tensor input_159_split_cast_fp16_0, tensor input_159_split_cast_fp16_1 = split(axis = input_159_split_axis_0, num_splits = input_159_split_num_splits_0, x = input_157_cast_fp16)[name = string("input_159_split_cast_fp16")]; - tensor input_159_split_1_sigmoid_cast_fp16 = sigmoid(x = input_159_split_cast_fp16_1)[name = string("input_159_split_1_sigmoid_cast_fp16")]; - tensor input_159_cast_fp16 = mul(x = input_159_split_cast_fp16_0, y = input_159_split_1_sigmoid_cast_fp16)[name = string("input_159_cast_fp16")]; - string input_161_pad_type_0 = const()[name = string("input_161_pad_type_0"), val = string("custom")]; - tensor input_161_pad_0 = const()[name = string("input_161_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_161_groups_0 = const()[name = string("input_161_groups_0"), val = int32(1024)]; - tensor input_161_strides_0 = const()[name = string("input_161_strides_0"), val = tensor([1, 1])]; - tensor input_161_dilations_0 = const()[name = string("input_161_dilations_0"), val = tensor([1, 1])]; - tensor const_278_to_fp16 = const()[name = string("const_278_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76376192)))]; - tensor const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76394688)))]; - tensor input_163_cast_fp16 = conv(bias = const_279_to_fp16, dilations = input_161_dilations_0, groups = input_161_groups_0, pad = input_161_pad_0, pad_type = input_161_pad_type_0, strides = input_161_strides_0, weight = const_278_to_fp16, x = input_159_cast_fp16)[name = string("input_163_cast_fp16")]; - tensor input_165_cast_fp16 = silu(x = input_163_cast_fp16)[name = string("input_165_cast_fp16")]; - string var_2336_pad_type_0 = const()[name = string("op_2336_pad_type_0"), val = string("valid")]; - tensor var_2336_strides_0 = const()[name = string("op_2336_strides_0"), val = tensor([1, 1])]; - tensor var_2336_pad_0 = const()[name = string("op_2336_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2336_dilations_0 = const()[name = string("op_2336_dilations_0"), val = tensor([1, 1])]; - int32 var_2336_groups_0 = const()[name = string("op_2336_groups_0"), val = int32(1)]; - tensor layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76396800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76790080))))[name = string("layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2336_cast_fp16 = conv(dilations = var_2336_dilations_0, groups = var_2336_groups_0, pad = var_2336_pad_0, pad_type = var_2336_pad_type_0, strides = var_2336_strides_0, weight = layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = string("op_2336_cast_fp16")]; - string var_2342_pad_type_0 = const()[name = string("op_2342_pad_type_0"), val = string("valid")]; - tensor var_2342_strides_0 = const()[name = string("op_2342_strides_0"), val = tensor([1, 1])]; - tensor var_2342_pad_0 = const()[name = string("op_2342_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2342_dilations_0 = const()[name = string("op_2342_dilations_0"), val = tensor([1, 1])]; - int32 var_2342_groups_0 = const()[name = string("op_2342_groups_0"), val = int32(1)]; - tensor layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76800128))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76791168))))[name = string("layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2342_cast_fp16 = conv(dilations = var_2342_dilations_0, groups = var_2342_groups_0, pad = var_2342_pad_0, pad_type = var_2342_pad_type_0, strides = var_2342_strides_0, weight = layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_165_cast_fp16)[name = string("op_2342_cast_fp16")]; - tensor x_35_cast_fp16 = add(x = var_2336_cast_fp16, y = var_2342_cast_fp16)[name = string("x_35_cast_fp16")]; - tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = x_35_cast_fp16)[name = string("inputs_57_cast_fp16")]; - tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; - fp16 var_2353_to_fp16 = const()[name = string("op_2353_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2353_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; - tensor input_167_gamma_0_to_fp16 = const()[name = string("input_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76931264)))]; - tensor input_167_beta_0_to_fp16 = const()[name = string("input_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76933376)))]; - fp16 input_167_epsilon_0_to_fp16 = const()[name = string("input_167_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_167_cast_fp16 = batch_norm(beta = input_167_beta_0_to_fp16, epsilon = input_167_epsilon_0_to_fp16, gamma = input_167_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("input_167_cast_fp16")]; - string var_2373_pad_type_0 = const()[name = string("op_2373_pad_type_0"), val = string("valid")]; - tensor var_2373_strides_0 = const()[name = string("op_2373_strides_0"), val = tensor([1, 1])]; - tensor var_2373_pad_0 = const()[name = string("op_2373_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2373_dilations_0 = const()[name = string("op_2373_dilations_0"), val = tensor([1, 1])]; - int32 var_2373_groups_0 = const()[name = string("op_2373_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76935488))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78508416))))[name = string("layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_2373_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2373_dilations_0, groups = var_2373_groups_0, pad = var_2373_pad_0, pad_type = var_2373_pad_type_0, strides = var_2373_strides_0, weight = layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = string("op_2373_cast_fp16")]; - string var_2379_pad_type_0 = const()[name = string("op_2379_pad_type_0"), val = string("valid")]; - tensor var_2379_strides_0 = const()[name = string("op_2379_strides_0"), val = tensor([1, 1])]; - tensor var_2379_pad_0 = const()[name = string("op_2379_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2379_dilations_0 = const()[name = string("op_2379_dilations_0"), val = tensor([1, 1])]; - int32 var_2379_groups_0 = const()[name = string("op_2379_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78557888))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78512576))))[name = string("layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2379_cast_fp16 = conv(dilations = var_2379_dilations_0, groups = var_2379_groups_0, pad = var_2379_pad_0, pad_type = var_2379_pad_type_0, strides = var_2379_strides_0, weight = layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_167_cast_fp16)[name = string("op_2379_cast_fp16")]; - tensor input_169_cast_fp16 = add(x = var_2373_cast_fp16, y = var_2379_cast_fp16)[name = string("input_169_cast_fp16")]; - tensor input_171_cast_fp16 = silu(x = input_169_cast_fp16)[name = string("input_171_cast_fp16")]; - string var_2390_pad_type_0 = const()[name = string("op_2390_pad_type_0"), val = string("valid")]; - tensor var_2390_strides_0 = const()[name = string("op_2390_strides_0"), val = tensor([1, 1])]; - tensor var_2390_pad_0 = const()[name = string("op_2390_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2390_dilations_0 = const()[name = string("op_2390_dilations_0"), val = tensor([1, 1])]; - int32 var_2390_groups_0 = const()[name = string("op_2390_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79082240))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80655168))))[name = string("layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2390_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2390_dilations_0, groups = var_2390_groups_0, pad = var_2390_pad_0, pad_type = var_2390_pad_type_0, strides = var_2390_strides_0, weight = layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = string("op_2390_cast_fp16")]; - string var_2396_pad_type_0 = const()[name = string("op_2396_pad_type_0"), val = string("valid")]; - tensor var_2396_strides_0 = const()[name = string("op_2396_strides_0"), val = tensor([1, 1])]; - tensor var_2396_pad_0 = const()[name = string("op_2396_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2396_dilations_0 = const()[name = string("op_2396_dilations_0"), val = tensor([1, 1])]; - int32 var_2396_groups_0 = const()[name = string("op_2396_groups_0"), val = int32(1)]; - tensor layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80719168))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80656256))))[name = string("layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2396_cast_fp16 = conv(dilations = var_2396_dilations_0, groups = var_2396_groups_0, pad = var_2396_pad_0, pad_type = var_2396_pad_type_0, strides = var_2396_strides_0, weight = layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_171_cast_fp16)[name = string("op_2396_cast_fp16")]; - tensor x_37_cast_fp16 = add(x = var_2390_cast_fp16, y = var_2396_cast_fp16)[name = string("x_37_cast_fp16")]; - fp16 var_2398_to_fp16 = const()[name = string("op_2398_to_fp16"), val = fp16(0x1p-1)]; - tensor var_2399_cast_fp16 = mul(x = x_37_cast_fp16, y = var_2398_to_fp16)[name = string("op_2399_cast_fp16")]; - tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = var_2399_cast_fp16)[name = string("inputs_59_cast_fp16")]; - tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; - fp16 var_2409_to_fp16 = const()[name = string("op_2409_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2409_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; - tensor inputs_61_gamma_0_to_fp16 = const()[name = string("inputs_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81243520)))]; - tensor inputs_61_beta_0_to_fp16 = const()[name = string("inputs_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81245632)))]; - fp16 inputs_61_epsilon_0_to_fp16 = const()[name = string("inputs_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_61_cast_fp16 = batch_norm(beta = inputs_61_beta_0_to_fp16, epsilon = inputs_61_epsilon_0_to_fp16, gamma = inputs_61_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("inputs_61_cast_fp16")]; - int32 var_2423 = const()[name = string("op_2423"), val = int32(3)]; - tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; - fp16 var_2454_to_fp16 = const()[name = string("op_2454_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2454_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; - tensor input_173_gamma_0_to_fp16 = const()[name = string("input_173_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81247744)))]; - tensor input_173_beta_0_to_fp16 = const()[name = string("input_173_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81249856)))]; - fp16 input_173_epsilon_0_to_fp16 = const()[name = string("input_173_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_173_cast_fp16 = batch_norm(beta = input_173_beta_0_to_fp16, epsilon = input_173_epsilon_0_to_fp16, gamma = input_173_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("input_173_cast_fp16")]; - string var_2474_pad_type_0 = const()[name = string("op_2474_pad_type_0"), val = string("valid")]; - tensor var_2474_strides_0 = const()[name = string("op_2474_strides_0"), val = tensor([1, 1])]; - tensor var_2474_pad_0 = const()[name = string("op_2474_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2474_dilations_0 = const()[name = string("op_2474_dilations_0"), val = tensor([1, 1])]; - int32 var_2474_groups_0 = const()[name = string("op_2474_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81251968))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82824896))))[name = string("layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_2474_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2474_dilations_0, groups = var_2474_groups_0, pad = var_2474_pad_0, pad_type = var_2474_pad_type_0, strides = var_2474_strides_0, weight = layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = string("op_2474_cast_fp16")]; - string var_2480_pad_type_0 = const()[name = string("op_2480_pad_type_0"), val = string("valid")]; - tensor var_2480_strides_0 = const()[name = string("op_2480_strides_0"), val = tensor([1, 1])]; - tensor var_2480_pad_0 = const()[name = string("op_2480_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2480_dilations_0 = const()[name = string("op_2480_dilations_0"), val = tensor([1, 1])]; - int32 var_2480_groups_0 = const()[name = string("op_2480_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82871040))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82829056))))[name = string("layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2480_cast_fp16 = conv(dilations = var_2480_dilations_0, groups = var_2480_groups_0, pad = var_2480_pad_0, pad_type = var_2480_pad_type_0, strides = var_2480_strides_0, weight = layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_173_cast_fp16)[name = string("op_2480_cast_fp16")]; - tensor input_175_cast_fp16 = add(x = var_2474_cast_fp16, y = var_2480_cast_fp16)[name = string("input_175_cast_fp16")]; - tensor input_177_cast_fp16 = silu(x = input_175_cast_fp16)[name = string("input_177_cast_fp16")]; - string var_2491_pad_type_0 = const()[name = string("op_2491_pad_type_0"), val = string("valid")]; - tensor var_2491_strides_0 = const()[name = string("op_2491_strides_0"), val = tensor([1, 1])]; - tensor var_2491_pad_0 = const()[name = string("op_2491_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2491_dilations_0 = const()[name = string("op_2491_dilations_0"), val = tensor([1, 1])]; - int32 var_2491_groups_0 = const()[name = string("op_2491_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83395392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84968320))))[name = string("layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2491_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2491_dilations_0, groups = var_2491_groups_0, pad = var_2491_pad_0, pad_type = var_2491_pad_type_0, strides = var_2491_strides_0, weight = layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = string("op_2491_cast_fp16")]; - string var_2497_pad_type_0 = const()[name = string("op_2497_pad_type_0"), val = string("valid")]; - tensor var_2497_strides_0 = const()[name = string("op_2497_strides_0"), val = tensor([1, 1])]; - tensor var_2497_pad_0 = const()[name = string("op_2497_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2497_dilations_0 = const()[name = string("op_2497_dilations_0"), val = tensor([1, 1])]; - int32 var_2497_groups_0 = const()[name = string("op_2497_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85048576))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84969408))))[name = string("layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2497_cast_fp16 = conv(dilations = var_2497_dilations_0, groups = var_2497_groups_0, pad = var_2497_pad_0, pad_type = var_2497_pad_type_0, strides = var_2497_strides_0, weight = layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_177_cast_fp16)[name = string("op_2497_cast_fp16")]; - tensor x_39_cast_fp16 = add(x = var_2491_cast_fp16, y = var_2497_cast_fp16)[name = string("x_39_cast_fp16")]; - fp16 var_2499_to_fp16 = const()[name = string("op_2499_to_fp16"), val = fp16(0x1p-1)]; - tensor var_2500_cast_fp16 = mul(x = x_39_cast_fp16, y = var_2499_to_fp16)[name = string("op_2500_cast_fp16")]; - tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = var_2500_cast_fp16)[name = string("inputs_63_cast_fp16")]; - tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; - fp16 var_2510_to_fp16 = const()[name = string("op_2510_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2510_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; - tensor obj_27_gamma_0_to_fp16 = const()[name = string("obj_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85572928)))]; - tensor obj_27_beta_0_to_fp16 = const()[name = string("obj_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85575040)))]; - fp16 obj_27_epsilon_0_to_fp16 = const()[name = string("obj_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_27_cast_fp16 = batch_norm(beta = obj_27_beta_0_to_fp16, epsilon = obj_27_epsilon_0_to_fp16, gamma = obj_27_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("obj_27_cast_fp16")]; - string var_2535_pad_type_0 = const()[name = string("op_2535_pad_type_0"), val = string("valid")]; - tensor var_2535_strides_0 = const()[name = string("op_2535_strides_0"), val = tensor([1, 1])]; - tensor var_2535_pad_0 = const()[name = string("op_2535_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2535_dilations_0 = const()[name = string("op_2535_dilations_0"), val = tensor([1, 1])]; - int32 var_2535_groups_0 = const()[name = string("op_2535_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85577152))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85970432))))[name = string("layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2535_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2535_dilations_0, groups = var_2535_groups_0, pad = var_2535_pad_0, pad_type = var_2535_pad_type_0, strides = var_2535_strides_0, weight = layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = string("op_2535_cast_fp16")]; - string var_2541_pad_type_0 = const()[name = string("op_2541_pad_type_0"), val = string("valid")]; - tensor var_2541_strides_0 = const()[name = string("op_2541_strides_0"), val = tensor([1, 1])]; - tensor var_2541_pad_0 = const()[name = string("op_2541_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2541_dilations_0 = const()[name = string("op_2541_dilations_0"), val = tensor([1, 1])]; - int32 var_2541_groups_0 = const()[name = string("op_2541_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85983168))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85971520))))[name = string("layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2541_cast_fp16 = conv(dilations = var_2541_dilations_0, groups = var_2541_groups_0, pad = var_2541_pad_0, pad_type = var_2541_pad_type_0, strides = var_2541_strides_0, weight = layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = string("op_2541_cast_fp16")]; - tensor query_25_cast_fp16 = add(x = var_2535_cast_fp16, y = var_2541_cast_fp16)[name = string("query_25_cast_fp16")]; - string var_2550_pad_type_0 = const()[name = string("op_2550_pad_type_0"), val = string("valid")]; - tensor var_2550_strides_0 = const()[name = string("op_2550_strides_0"), val = tensor([1, 1])]; - tensor var_2550_pad_0 = const()[name = string("op_2550_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2550_dilations_0 = const()[name = string("op_2550_dilations_0"), val = tensor([1, 1])]; - int32 var_2550_groups_0 = const()[name = string("op_2550_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86114304))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86507584))))[name = string("layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2550_cast_fp16 = conv(dilations = var_2550_dilations_0, groups = var_2550_groups_0, pad = var_2550_pad_0, pad_type = var_2550_pad_type_0, strides = var_2550_strides_0, weight = layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = string("op_2550_cast_fp16")]; - string var_2556_pad_type_0 = const()[name = string("op_2556_pad_type_0"), val = string("valid")]; - tensor var_2556_strides_0 = const()[name = string("op_2556_strides_0"), val = tensor([1, 1])]; - tensor var_2556_pad_0 = const()[name = string("op_2556_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2556_dilations_0 = const()[name = string("op_2556_dilations_0"), val = tensor([1, 1])]; - int32 var_2556_groups_0 = const()[name = string("op_2556_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86524800))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86508672))))[name = string("layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2556_cast_fp16 = conv(dilations = var_2556_dilations_0, groups = var_2556_groups_0, pad = var_2556_pad_0, pad_type = var_2556_pad_type_0, strides = var_2556_strides_0, weight = layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = string("op_2556_cast_fp16")]; - tensor key_13_cast_fp16 = add(x = var_2550_cast_fp16, y = var_2556_cast_fp16)[name = string("key_13_cast_fp16")]; - string var_2566_pad_type_0 = const()[name = string("op_2566_pad_type_0"), val = string("valid")]; - tensor var_2566_strides_0 = const()[name = string("op_2566_strides_0"), val = tensor([1, 1])]; - tensor var_2566_pad_0 = const()[name = string("op_2566_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2566_dilations_0 = const()[name = string("op_2566_dilations_0"), val = tensor([1, 1])]; - int32 var_2566_groups_0 = const()[name = string("op_2566_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86655936))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87049216))))[name = string("layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2566_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2566_dilations_0, groups = var_2566_groups_0, pad = var_2566_pad_0, pad_type = var_2566_pad_type_0, strides = var_2566_strides_0, weight = layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = string("op_2566_cast_fp16")]; - string var_2572_pad_type_0 = const()[name = string("op_2572_pad_type_0"), val = string("valid")]; - tensor var_2572_strides_0 = const()[name = string("op_2572_strides_0"), val = tensor([1, 1])]; - tensor var_2572_pad_0 = const()[name = string("op_2572_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2572_dilations_0 = const()[name = string("op_2572_dilations_0"), val = tensor([1, 1])]; - int32 var_2572_groups_0 = const()[name = string("op_2572_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87059328))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87050304))))[name = string("layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2572_cast_fp16 = conv(dilations = var_2572_dilations_0, groups = var_2572_groups_0, pad = var_2572_pad_0, pad_type = var_2572_pad_type_0, strides = var_2572_strides_0, weight = layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = string("op_2572_cast_fp16")]; - tensor value_13_cast_fp16 = add(x = var_2566_cast_fp16, y = var_2572_cast_fp16)[name = string("value_13_cast_fp16")]; - tensor var_2575_to_fp16 = const()[name = string("op_2575_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87190464)))]; - tensor query_27_cast_fp16 = add(x = query_25_cast_fp16, y = var_2575_to_fp16)[name = string("query_27_cast_fp16")]; - tensor var_2578_to_fp16 = const()[name = string("op_2578_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87192576)))]; - tensor q_with_bias_v_13_cast_fp16 = add(x = query_25_cast_fp16, y = var_2578_to_fp16)[name = string("q_with_bias_v_13_cast_fp16")]; - string var_2588_pad_type_0 = const()[name = string("op_2588_pad_type_0"), val = string("valid")]; - tensor var_2588_strides_0 = const()[name = string("op_2588_strides_0"), val = tensor([1, 1])]; - tensor var_2588_pad_0 = const()[name = string("op_2588_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2588_dilations_0 = const()[name = string("op_2588_dilations_0"), val = tensor([1, 1])]; - int32 var_2588_groups_0 = const()[name = string("op_2588_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87194688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87587968))))[name = string("layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_2588_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2588_dilations_0, groups = var_2588_groups_0, pad = var_2588_pad_0, pad_type = var_2588_pad_type_0, strides = var_2588_strides_0, weight = layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_2588_cast_fp16")]; - string var_2594_pad_type_0 = const()[name = string("op_2594_pad_type_0"), val = string("valid")]; - tensor var_2594_strides_0 = const()[name = string("op_2594_strides_0"), val = tensor([1, 1])]; - tensor var_2594_pad_0 = const()[name = string("op_2594_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2594_dilations_0 = const()[name = string("op_2594_dilations_0"), val = tensor([1, 1])]; - int32 var_2594_groups_0 = const()[name = string("op_2594_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87627712))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87589056))))[name = string("layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2594_cast_fp16 = conv(dilations = var_2594_dilations_0, groups = var_2594_groups_0, pad = var_2594_pad_0, pad_type = var_2594_pad_type_0, strides = var_2594_strides_0, weight = layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_2594_cast_fp16")]; - tensor p_13_cast_fp16 = add(x = var_2588_cast_fp16, y = var_2594_cast_fp16)[name = string("p_13_cast_fp16")]; - tensor var_2598 = const()[name = string("op_2598"), val = tensor([1, 8, 128, 188])]; - tensor var_2599_cast_fp16 = reshape(shape = var_2598, x = q_with_bias_v_13_cast_fp16)[name = string("op_2599_cast_fp16")]; - tensor var_2600 = const()[name = string("op_2600"), val = tensor([1, 8, 128, -1])]; - tensor var_2601_cast_fp16 = reshape(shape = var_2600, x = p_13_cast_fp16)[name = string("op_2601_cast_fp16")]; - bool matrix_bd_49_transpose_x_0 = const()[name = string("matrix_bd_49_transpose_x_0"), val = bool(true)]; - bool matrix_bd_49_transpose_y_0 = const()[name = string("matrix_bd_49_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_49_cast_fp16 = matmul(transpose_x = matrix_bd_49_transpose_x_0, transpose_y = matrix_bd_49_transpose_y_0, x = var_2599_cast_fp16, y = var_2601_cast_fp16)[name = string("matrix_bd_49_cast_fp16")]; - tensor matrix_bd_51_pad_0 = const()[name = string("matrix_bd_51_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_51_mode_0 = const()[name = string("matrix_bd_51_mode_0"), val = string("constant")]; - fp16 const_76_to_fp16 = const()[name = string("const_76_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_51_cast_fp16 = pad(constant_val = const_76_to_fp16, mode = matrix_bd_51_mode_0, pad = matrix_bd_51_pad_0, x = matrix_bd_49_cast_fp16)[name = string("matrix_bd_51_cast_fp16")]; - tensor var_2610 = const()[name = string("op_2610"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_53_cast_fp16 = reshape(shape = var_2610, x = matrix_bd_51_cast_fp16)[name = string("matrix_bd_53_cast_fp16")]; - tensor var_2614_begin_0 = const()[name = string("op_2614_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_2614_end_0 = const()[name = string("op_2614_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_2614_end_mask_0 = const()[name = string("op_2614_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_2614_cast_fp16 = slice_by_index(begin = var_2614_begin_0, end = var_2614_end_0, end_mask = var_2614_end_mask_0, x = matrix_bd_53_cast_fp16)[name = string("op_2614_cast_fp16")]; - tensor var_2615 = const()[name = string("op_2615"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_55_cast_fp16 = reshape(shape = var_2615, x = var_2614_cast_fp16)[name = string("matrix_bd_55_cast_fp16")]; - tensor var_2620_begin_0 = const()[name = string("op_2620_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2620_end_0 = const()[name = string("op_2620_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_2620_end_mask_0 = const()[name = string("op_2620_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_2620_cast_fp16 = slice_by_index(begin = var_2620_begin_0, end = var_2620_end_0, end_mask = var_2620_end_mask_0, x = matrix_bd_55_cast_fp16)[name = string("op_2620_cast_fp16")]; - fp16 var_2621_to_fp16 = const()[name = string("op_2621_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_13_cast_fp16 = mul(x = var_2620_cast_fp16, y = var_2621_to_fp16)[name = string("qk_mask_13_cast_fp16")]; - tensor var_2625 = const()[name = string("op_2625"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_13_cast_fp16 = reshape(shape = var_2625, x = query_27_cast_fp16)[name = string("mh_q_13_cast_fp16")]; - fp16 var_2627_to_fp16 = const()[name = string("op_2627_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_2628_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_2627_to_fp16)[name = string("op_2628_cast_fp16")]; - tensor var_2631 = const()[name = string("op_2631"), val = tensor([1, 8, 128, 188])]; - tensor var_2632_cast_fp16 = reshape(shape = var_2631, x = key_13_cast_fp16)[name = string("op_2632_cast_fp16")]; - bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; - bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; - tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_2628_cast_fp16, y = var_2632_cast_fp16)[name = string("mh_w_25_cast_fp16")]; - tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = qk_mask_13_cast_fp16)[name = string("mh_w_27_cast_fp16")]; - tensor var_2636_cast_fp16 = softmax(axis = var_2423, x = mh_w_27_cast_fp16)[name = string("op_2636_cast_fp16")]; - tensor var_2637 = const()[name = string("op_2637"), val = tensor([1, 8, 128, 188])]; - tensor var_2638_cast_fp16 = reshape(shape = var_2637, x = value_13_cast_fp16)[name = string("op_2638_cast_fp16")]; - bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; - bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; - tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_2638_cast_fp16, y = var_2636_cast_fp16)[name = string("attn_13_cast_fp16")]; - tensor var_2641 = const()[name = string("op_2641"), val = tensor([1, 1024, 1, 188])]; - tensor input_179_cast_fp16 = reshape(shape = var_2641, x = attn_13_cast_fp16)[name = string("input_179_cast_fp16")]; - string var_2651_pad_type_0 = const()[name = string("op_2651_pad_type_0"), val = string("valid")]; - tensor var_2651_strides_0 = const()[name = string("op_2651_strides_0"), val = tensor([1, 1])]; - tensor var_2651_pad_0 = const()[name = string("op_2651_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2651_dilations_0 = const()[name = string("op_2651_dilations_0"), val = tensor([1, 1])]; - int32 var_2651_groups_0 = const()[name = string("op_2651_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87758848))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88152128))))[name = string("layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2651_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2651_dilations_0, groups = var_2651_groups_0, pad = var_2651_pad_0, pad_type = var_2651_pad_type_0, strides = var_2651_strides_0, weight = layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = string("op_2651_cast_fp16")]; - string var_2657_pad_type_0 = const()[name = string("op_2657_pad_type_0"), val = string("valid")]; - tensor var_2657_strides_0 = const()[name = string("op_2657_strides_0"), val = tensor([1, 1])]; - tensor var_2657_pad_0 = const()[name = string("op_2657_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2657_dilations_0 = const()[name = string("op_2657_dilations_0"), val = tensor([1, 1])]; - int32 var_2657_groups_0 = const()[name = string("op_2657_groups_0"), val = int32(1)]; - tensor layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88161408))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88153216))))[name = string("layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2657_cast_fp16 = conv(dilations = var_2657_dilations_0, groups = var_2657_groups_0, pad = var_2657_pad_0, pad_type = var_2657_pad_type_0, strides = var_2657_strides_0, weight = layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_179_cast_fp16)[name = string("op_2657_cast_fp16")]; - tensor obj_29_cast_fp16 = add(x = var_2651_cast_fp16, y = var_2657_cast_fp16)[name = string("obj_29_cast_fp16")]; - tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_29_cast_fp16)[name = string("inputs_65_cast_fp16")]; - tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; - fp16 var_2668_to_fp16 = const()[name = string("op_2668_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2668_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; - tensor input_181_gamma_0_to_fp16 = const()[name = string("input_181_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88292544)))]; - tensor input_181_beta_0_to_fp16 = const()[name = string("input_181_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88294656)))]; - fp16 input_181_epsilon_0_to_fp16 = const()[name = string("input_181_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_181_cast_fp16 = batch_norm(beta = input_181_beta_0_to_fp16, epsilon = input_181_epsilon_0_to_fp16, gamma = input_181_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("input_181_cast_fp16")]; - string var_2689_pad_type_0 = const()[name = string("op_2689_pad_type_0"), val = string("valid")]; - tensor var_2689_strides_0 = const()[name = string("op_2689_strides_0"), val = tensor([1, 1])]; - tensor var_2689_pad_0 = const()[name = string("op_2689_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2689_dilations_0 = const()[name = string("op_2689_dilations_0"), val = tensor([1, 1])]; - int32 var_2689_groups_0 = const()[name = string("op_2689_groups_0"), val = int32(1)]; - tensor layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88296768))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89083264))))[name = string("layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_2689_cast_fp16 = conv(dilations = var_2689_dilations_0, groups = var_2689_groups_0, pad = var_2689_pad_0, pad_type = var_2689_pad_type_0, strides = var_2689_strides_0, weight = layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = string("op_2689_cast_fp16")]; - string var_2695_pad_type_0 = const()[name = string("op_2695_pad_type_0"), val = string("valid")]; - tensor var_2695_strides_0 = const()[name = string("op_2695_strides_0"), val = tensor([1, 1])]; - tensor var_2695_pad_0 = const()[name = string("op_2695_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2695_dilations_0 = const()[name = string("op_2695_dilations_0"), val = tensor([1, 1])]; - int32 var_2695_groups_0 = const()[name = string("op_2695_groups_0"), val = int32(1)]; - tensor layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89105920))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89085376))))[name = string("layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2695_cast_fp16 = conv(dilations = var_2695_dilations_0, groups = var_2695_groups_0, pad = var_2695_pad_0, pad_type = var_2695_pad_type_0, strides = var_2695_strides_0, weight = layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_181_cast_fp16)[name = string("op_2695_cast_fp16")]; - tensor input_183_cast_fp16 = add(x = var_2689_cast_fp16, y = var_2695_cast_fp16)[name = string("input_183_cast_fp16")]; - int32 input_185_split_num_splits_0 = const()[name = string("input_185_split_num_splits_0"), val = int32(2)]; - int32 input_185_split_axis_0 = const()[name = string("input_185_split_axis_0"), val = int32(1)]; - tensor input_185_split_cast_fp16_0, tensor input_185_split_cast_fp16_1 = split(axis = input_185_split_axis_0, num_splits = input_185_split_num_splits_0, x = input_183_cast_fp16)[name = string("input_185_split_cast_fp16")]; - tensor input_185_split_1_sigmoid_cast_fp16 = sigmoid(x = input_185_split_cast_fp16_1)[name = string("input_185_split_1_sigmoid_cast_fp16")]; - tensor input_185_cast_fp16 = mul(x = input_185_split_cast_fp16_0, y = input_185_split_1_sigmoid_cast_fp16)[name = string("input_185_cast_fp16")]; - string input_187_pad_type_0 = const()[name = string("input_187_pad_type_0"), val = string("custom")]; - tensor input_187_pad_0 = const()[name = string("input_187_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_187_groups_0 = const()[name = string("input_187_groups_0"), val = int32(1024)]; - tensor input_187_strides_0 = const()[name = string("input_187_strides_0"), val = tensor([1, 1])]; - tensor input_187_dilations_0 = const()[name = string("input_187_dilations_0"), val = tensor([1, 1])]; - tensor const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89368128)))]; - tensor const_281_to_fp16 = const()[name = string("const_281_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89386624)))]; - tensor input_189_cast_fp16 = conv(bias = const_281_to_fp16, dilations = input_187_dilations_0, groups = input_187_groups_0, pad = input_187_pad_0, pad_type = input_187_pad_type_0, strides = input_187_strides_0, weight = const_280_to_fp16, x = input_185_cast_fp16)[name = string("input_189_cast_fp16")]; - tensor input_191_cast_fp16 = silu(x = input_189_cast_fp16)[name = string("input_191_cast_fp16")]; - string var_2717_pad_type_0 = const()[name = string("op_2717_pad_type_0"), val = string("valid")]; - tensor var_2717_strides_0 = const()[name = string("op_2717_strides_0"), val = tensor([1, 1])]; - tensor var_2717_pad_0 = const()[name = string("op_2717_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2717_dilations_0 = const()[name = string("op_2717_dilations_0"), val = tensor([1, 1])]; - int32 var_2717_groups_0 = const()[name = string("op_2717_groups_0"), val = int32(1)]; - tensor layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89388736))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89782016))))[name = string("layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2717_cast_fp16 = conv(dilations = var_2717_dilations_0, groups = var_2717_groups_0, pad = var_2717_pad_0, pad_type = var_2717_pad_type_0, strides = var_2717_strides_0, weight = layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = string("op_2717_cast_fp16")]; - string var_2723_pad_type_0 = const()[name = string("op_2723_pad_type_0"), val = string("valid")]; - tensor var_2723_strides_0 = const()[name = string("op_2723_strides_0"), val = tensor([1, 1])]; - tensor var_2723_pad_0 = const()[name = string("op_2723_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2723_dilations_0 = const()[name = string("op_2723_dilations_0"), val = tensor([1, 1])]; - int32 var_2723_groups_0 = const()[name = string("op_2723_groups_0"), val = int32(1)]; - tensor layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89792128))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89783104))))[name = string("layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2723_cast_fp16 = conv(dilations = var_2723_dilations_0, groups = var_2723_groups_0, pad = var_2723_pad_0, pad_type = var_2723_pad_type_0, strides = var_2723_strides_0, weight = layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_191_cast_fp16)[name = string("op_2723_cast_fp16")]; - tensor x_41_cast_fp16 = add(x = var_2717_cast_fp16, y = var_2723_cast_fp16)[name = string("x_41_cast_fp16")]; - tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = x_41_cast_fp16)[name = string("inputs_67_cast_fp16")]; - tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; - fp16 var_2734_to_fp16 = const()[name = string("op_2734_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2734_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; - tensor input_193_gamma_0_to_fp16 = const()[name = string("input_193_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89923264)))]; - tensor input_193_beta_0_to_fp16 = const()[name = string("input_193_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89925376)))]; - fp16 input_193_epsilon_0_to_fp16 = const()[name = string("input_193_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_193_cast_fp16 = batch_norm(beta = input_193_beta_0_to_fp16, epsilon = input_193_epsilon_0_to_fp16, gamma = input_193_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("input_193_cast_fp16")]; - string var_2754_pad_type_0 = const()[name = string("op_2754_pad_type_0"), val = string("valid")]; - tensor var_2754_strides_0 = const()[name = string("op_2754_strides_0"), val = tensor([1, 1])]; - tensor var_2754_pad_0 = const()[name = string("op_2754_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2754_dilations_0 = const()[name = string("op_2754_dilations_0"), val = tensor([1, 1])]; - int32 var_2754_groups_0 = const()[name = string("op_2754_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89927488))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91500416))))[name = string("layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_2754_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2754_dilations_0, groups = var_2754_groups_0, pad = var_2754_pad_0, pad_type = var_2754_pad_type_0, strides = var_2754_strides_0, weight = layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = string("op_2754_cast_fp16")]; - string var_2760_pad_type_0 = const()[name = string("op_2760_pad_type_0"), val = string("valid")]; - tensor var_2760_strides_0 = const()[name = string("op_2760_strides_0"), val = tensor([1, 1])]; - tensor var_2760_pad_0 = const()[name = string("op_2760_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2760_dilations_0 = const()[name = string("op_2760_dilations_0"), val = tensor([1, 1])]; - int32 var_2760_groups_0 = const()[name = string("op_2760_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91547392))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91504576))))[name = string("layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2760_cast_fp16 = conv(dilations = var_2760_dilations_0, groups = var_2760_groups_0, pad = var_2760_pad_0, pad_type = var_2760_pad_type_0, strides = var_2760_strides_0, weight = layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_193_cast_fp16)[name = string("op_2760_cast_fp16")]; - tensor input_195_cast_fp16 = add(x = var_2754_cast_fp16, y = var_2760_cast_fp16)[name = string("input_195_cast_fp16")]; - tensor input_197_cast_fp16 = silu(x = input_195_cast_fp16)[name = string("input_197_cast_fp16")]; - string var_2771_pad_type_0 = const()[name = string("op_2771_pad_type_0"), val = string("valid")]; - tensor var_2771_strides_0 = const()[name = string("op_2771_strides_0"), val = tensor([1, 1])]; - tensor var_2771_pad_0 = const()[name = string("op_2771_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2771_dilations_0 = const()[name = string("op_2771_dilations_0"), val = tensor([1, 1])]; - int32 var_2771_groups_0 = const()[name = string("op_2771_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92071744))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93644672))))[name = string("layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2771_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2771_dilations_0, groups = var_2771_groups_0, pad = var_2771_pad_0, pad_type = var_2771_pad_type_0, strides = var_2771_strides_0, weight = layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = string("op_2771_cast_fp16")]; - string var_2777_pad_type_0 = const()[name = string("op_2777_pad_type_0"), val = string("valid")]; - tensor var_2777_strides_0 = const()[name = string("op_2777_strides_0"), val = tensor([1, 1])]; - tensor var_2777_pad_0 = const()[name = string("op_2777_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2777_dilations_0 = const()[name = string("op_2777_dilations_0"), val = tensor([1, 1])]; - int32 var_2777_groups_0 = const()[name = string("op_2777_groups_0"), val = int32(1)]; - tensor layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93704640))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93645760))))[name = string("layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2777_cast_fp16 = conv(dilations = var_2777_dilations_0, groups = var_2777_groups_0, pad = var_2777_pad_0, pad_type = var_2777_pad_type_0, strides = var_2777_strides_0, weight = layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_197_cast_fp16)[name = string("op_2777_cast_fp16")]; - tensor x_43_cast_fp16 = add(x = var_2771_cast_fp16, y = var_2777_cast_fp16)[name = string("x_43_cast_fp16")]; - fp16 var_2779_to_fp16 = const()[name = string("op_2779_to_fp16"), val = fp16(0x1p-1)]; - tensor var_2780_cast_fp16 = mul(x = x_43_cast_fp16, y = var_2779_to_fp16)[name = string("op_2780_cast_fp16")]; - tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = var_2780_cast_fp16)[name = string("inputs_69_cast_fp16")]; - tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; - fp16 var_2790_to_fp16 = const()[name = string("op_2790_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2790_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; - tensor inputs_71_gamma_0_to_fp16 = const()[name = string("inputs_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94228992)))]; - tensor inputs_71_beta_0_to_fp16 = const()[name = string("inputs_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94231104)))]; - fp16 inputs_71_epsilon_0_to_fp16 = const()[name = string("inputs_71_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_71_cast_fp16 = batch_norm(beta = inputs_71_beta_0_to_fp16, epsilon = inputs_71_epsilon_0_to_fp16, gamma = inputs_71_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("inputs_71_cast_fp16")]; - int32 var_2804 = const()[name = string("op_2804"), val = int32(3)]; - tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; - fp16 var_2835_to_fp16 = const()[name = string("op_2835_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2835_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; - tensor input_199_gamma_0_to_fp16 = const()[name = string("input_199_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94233216)))]; - tensor input_199_beta_0_to_fp16 = const()[name = string("input_199_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94235328)))]; - fp16 input_199_epsilon_0_to_fp16 = const()[name = string("input_199_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_199_cast_fp16 = batch_norm(beta = input_199_beta_0_to_fp16, epsilon = input_199_epsilon_0_to_fp16, gamma = input_199_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_199_cast_fp16")]; - string var_2855_pad_type_0 = const()[name = string("op_2855_pad_type_0"), val = string("valid")]; - tensor var_2855_strides_0 = const()[name = string("op_2855_strides_0"), val = tensor([1, 1])]; - tensor var_2855_pad_0 = const()[name = string("op_2855_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2855_dilations_0 = const()[name = string("op_2855_dilations_0"), val = tensor([1, 1])]; - int32 var_2855_groups_0 = const()[name = string("op_2855_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94237440))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95810368))))[name = string("layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_2855_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2855_dilations_0, groups = var_2855_groups_0, pad = var_2855_pad_0, pad_type = var_2855_pad_type_0, strides = var_2855_strides_0, weight = layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = string("op_2855_cast_fp16")]; - string var_2861_pad_type_0 = const()[name = string("op_2861_pad_type_0"), val = string("valid")]; - tensor var_2861_strides_0 = const()[name = string("op_2861_strides_0"), val = tensor([1, 1])]; - tensor var_2861_pad_0 = const()[name = string("op_2861_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2861_dilations_0 = const()[name = string("op_2861_dilations_0"), val = tensor([1, 1])]; - int32 var_2861_groups_0 = const()[name = string("op_2861_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95855040))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95814528))))[name = string("layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2861_cast_fp16 = conv(dilations = var_2861_dilations_0, groups = var_2861_groups_0, pad = var_2861_pad_0, pad_type = var_2861_pad_type_0, strides = var_2861_strides_0, weight = layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_199_cast_fp16)[name = string("op_2861_cast_fp16")]; - tensor input_201_cast_fp16 = add(x = var_2855_cast_fp16, y = var_2861_cast_fp16)[name = string("input_201_cast_fp16")]; - tensor input_203_cast_fp16 = silu(x = input_201_cast_fp16)[name = string("input_203_cast_fp16")]; - string var_2872_pad_type_0 = const()[name = string("op_2872_pad_type_0"), val = string("valid")]; - tensor var_2872_strides_0 = const()[name = string("op_2872_strides_0"), val = tensor([1, 1])]; - tensor var_2872_pad_0 = const()[name = string("op_2872_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2872_dilations_0 = const()[name = string("op_2872_dilations_0"), val = tensor([1, 1])]; - int32 var_2872_groups_0 = const()[name = string("op_2872_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96379392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97952320))))[name = string("layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_2872_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2872_dilations_0, groups = var_2872_groups_0, pad = var_2872_pad_0, pad_type = var_2872_pad_type_0, strides = var_2872_strides_0, weight = layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = string("op_2872_cast_fp16")]; - string var_2878_pad_type_0 = const()[name = string("op_2878_pad_type_0"), val = string("valid")]; - tensor var_2878_strides_0 = const()[name = string("op_2878_strides_0"), val = tensor([1, 1])]; - tensor var_2878_pad_0 = const()[name = string("op_2878_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2878_dilations_0 = const()[name = string("op_2878_dilations_0"), val = tensor([1, 1])]; - int32 var_2878_groups_0 = const()[name = string("op_2878_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98029120))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97953408))))[name = string("layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2878_cast_fp16 = conv(dilations = var_2878_dilations_0, groups = var_2878_groups_0, pad = var_2878_pad_0, pad_type = var_2878_pad_type_0, strides = var_2878_strides_0, weight = layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_203_cast_fp16)[name = string("op_2878_cast_fp16")]; - tensor x_45_cast_fp16 = add(x = var_2872_cast_fp16, y = var_2878_cast_fp16)[name = string("x_45_cast_fp16")]; - fp16 var_2880_to_fp16 = const()[name = string("op_2880_to_fp16"), val = fp16(0x1p-1)]; - tensor var_2881_cast_fp16 = mul(x = x_45_cast_fp16, y = var_2880_to_fp16)[name = string("op_2881_cast_fp16")]; - tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = var_2881_cast_fp16)[name = string("inputs_73_cast_fp16")]; - tensor out_73_axes_0 = const()[name = string("out_73_axes_0"), val = tensor([1])]; - fp16 var_2891_to_fp16 = const()[name = string("op_2891_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2891_to_fp16, x = inputs_73_cast_fp16)[name = string("out_73_cast_fp16")]; - tensor obj_31_gamma_0_to_fp16 = const()[name = string("obj_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98553472)))]; - tensor obj_31_beta_0_to_fp16 = const()[name = string("obj_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98555584)))]; - fp16 obj_31_epsilon_0_to_fp16 = const()[name = string("obj_31_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_31_cast_fp16 = batch_norm(beta = obj_31_beta_0_to_fp16, epsilon = obj_31_epsilon_0_to_fp16, gamma = obj_31_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_73_cast_fp16)[name = string("obj_31_cast_fp16")]; - string var_2916_pad_type_0 = const()[name = string("op_2916_pad_type_0"), val = string("valid")]; - tensor var_2916_strides_0 = const()[name = string("op_2916_strides_0"), val = tensor([1, 1])]; - tensor var_2916_pad_0 = const()[name = string("op_2916_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2916_dilations_0 = const()[name = string("op_2916_dilations_0"), val = tensor([1, 1])]; - int32 var_2916_groups_0 = const()[name = string("op_2916_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98557696))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98950976))))[name = string("layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2916_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2916_dilations_0, groups = var_2916_groups_0, pad = var_2916_pad_0, pad_type = var_2916_pad_type_0, strides = var_2916_strides_0, weight = layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = string("op_2916_cast_fp16")]; - string var_2922_pad_type_0 = const()[name = string("op_2922_pad_type_0"), val = string("valid")]; - tensor var_2922_strides_0 = const()[name = string("op_2922_strides_0"), val = tensor([1, 1])]; - tensor var_2922_pad_0 = const()[name = string("op_2922_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2922_dilations_0 = const()[name = string("op_2922_dilations_0"), val = tensor([1, 1])]; - int32 var_2922_groups_0 = const()[name = string("op_2922_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98963776))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98952064))))[name = string("layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2922_cast_fp16 = conv(dilations = var_2922_dilations_0, groups = var_2922_groups_0, pad = var_2922_pad_0, pad_type = var_2922_pad_type_0, strides = var_2922_strides_0, weight = layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = string("op_2922_cast_fp16")]; - tensor query_29_cast_fp16 = add(x = var_2916_cast_fp16, y = var_2922_cast_fp16)[name = string("query_29_cast_fp16")]; - string var_2931_pad_type_0 = const()[name = string("op_2931_pad_type_0"), val = string("valid")]; - tensor var_2931_strides_0 = const()[name = string("op_2931_strides_0"), val = tensor([1, 1])]; - tensor var_2931_pad_0 = const()[name = string("op_2931_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2931_dilations_0 = const()[name = string("op_2931_dilations_0"), val = tensor([1, 1])]; - int32 var_2931_groups_0 = const()[name = string("op_2931_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99094912))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99488192))))[name = string("layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2931_cast_fp16 = conv(dilations = var_2931_dilations_0, groups = var_2931_groups_0, pad = var_2931_pad_0, pad_type = var_2931_pad_type_0, strides = var_2931_strides_0, weight = layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = string("op_2931_cast_fp16")]; - string var_2937_pad_type_0 = const()[name = string("op_2937_pad_type_0"), val = string("valid")]; - tensor var_2937_strides_0 = const()[name = string("op_2937_strides_0"), val = tensor([1, 1])]; - tensor var_2937_pad_0 = const()[name = string("op_2937_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2937_dilations_0 = const()[name = string("op_2937_dilations_0"), val = tensor([1, 1])]; - int32 var_2937_groups_0 = const()[name = string("op_2937_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99507264))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99489280))))[name = string("layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2937_cast_fp16 = conv(dilations = var_2937_dilations_0, groups = var_2937_groups_0, pad = var_2937_pad_0, pad_type = var_2937_pad_type_0, strides = var_2937_strides_0, weight = layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = string("op_2937_cast_fp16")]; - tensor key_15_cast_fp16 = add(x = var_2931_cast_fp16, y = var_2937_cast_fp16)[name = string("key_15_cast_fp16")]; - string var_2947_pad_type_0 = const()[name = string("op_2947_pad_type_0"), val = string("valid")]; - tensor var_2947_strides_0 = const()[name = string("op_2947_strides_0"), val = tensor([1, 1])]; - tensor var_2947_pad_0 = const()[name = string("op_2947_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2947_dilations_0 = const()[name = string("op_2947_dilations_0"), val = tensor([1, 1])]; - int32 var_2947_groups_0 = const()[name = string("op_2947_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99638400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100031680))))[name = string("layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_2947_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2947_dilations_0, groups = var_2947_groups_0, pad = var_2947_pad_0, pad_type = var_2947_pad_type_0, strides = var_2947_strides_0, weight = layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = string("op_2947_cast_fp16")]; - string var_2953_pad_type_0 = const()[name = string("op_2953_pad_type_0"), val = string("valid")]; - tensor var_2953_strides_0 = const()[name = string("op_2953_strides_0"), val = tensor([1, 1])]; - tensor var_2953_pad_0 = const()[name = string("op_2953_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2953_dilations_0 = const()[name = string("op_2953_dilations_0"), val = tensor([1, 1])]; - int32 var_2953_groups_0 = const()[name = string("op_2953_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100043072))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100032768))))[name = string("layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2953_cast_fp16 = conv(dilations = var_2953_dilations_0, groups = var_2953_groups_0, pad = var_2953_pad_0, pad_type = var_2953_pad_type_0, strides = var_2953_strides_0, weight = layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = string("op_2953_cast_fp16")]; - tensor value_15_cast_fp16 = add(x = var_2947_cast_fp16, y = var_2953_cast_fp16)[name = string("value_15_cast_fp16")]; - tensor var_2956_to_fp16 = const()[name = string("op_2956_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100174208)))]; - tensor query_31_cast_fp16 = add(x = query_29_cast_fp16, y = var_2956_to_fp16)[name = string("query_31_cast_fp16")]; - tensor var_2959_to_fp16 = const()[name = string("op_2959_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100176320)))]; - tensor q_with_bias_v_15_cast_fp16 = add(x = query_29_cast_fp16, y = var_2959_to_fp16)[name = string("q_with_bias_v_15_cast_fp16")]; - string var_2969_pad_type_0 = const()[name = string("op_2969_pad_type_0"), val = string("valid")]; - tensor var_2969_strides_0 = const()[name = string("op_2969_strides_0"), val = tensor([1, 1])]; - tensor var_2969_pad_0 = const()[name = string("op_2969_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2969_dilations_0 = const()[name = string("op_2969_dilations_0"), val = tensor([1, 1])]; - int32 var_2969_groups_0 = const()[name = string("op_2969_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100178432))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100571712))))[name = string("layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_2969_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2969_dilations_0, groups = var_2969_groups_0, pad = var_2969_pad_0, pad_type = var_2969_pad_type_0, strides = var_2969_strides_0, weight = layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_2969_cast_fp16")]; - string var_2975_pad_type_0 = const()[name = string("op_2975_pad_type_0"), val = string("valid")]; - tensor var_2975_strides_0 = const()[name = string("op_2975_strides_0"), val = tensor([1, 1])]; - tensor var_2975_pad_0 = const()[name = string("op_2975_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_2975_dilations_0 = const()[name = string("op_2975_dilations_0"), val = tensor([1, 1])]; - int32 var_2975_groups_0 = const()[name = string("op_2975_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100607552))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100572800))))[name = string("layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_2975_cast_fp16 = conv(dilations = var_2975_dilations_0, groups = var_2975_groups_0, pad = var_2975_pad_0, pad_type = var_2975_pad_type_0, strides = var_2975_strides_0, weight = layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_2975_cast_fp16")]; - tensor p_15_cast_fp16 = add(x = var_2969_cast_fp16, y = var_2975_cast_fp16)[name = string("p_15_cast_fp16")]; - tensor var_2979 = const()[name = string("op_2979"), val = tensor([1, 8, 128, 188])]; - tensor var_2980_cast_fp16 = reshape(shape = var_2979, x = q_with_bias_v_15_cast_fp16)[name = string("op_2980_cast_fp16")]; - tensor var_2981 = const()[name = string("op_2981"), val = tensor([1, 8, 128, -1])]; - tensor var_2982_cast_fp16 = reshape(shape = var_2981, x = p_15_cast_fp16)[name = string("op_2982_cast_fp16")]; - bool matrix_bd_57_transpose_x_0 = const()[name = string("matrix_bd_57_transpose_x_0"), val = bool(true)]; - bool matrix_bd_57_transpose_y_0 = const()[name = string("matrix_bd_57_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_57_cast_fp16 = matmul(transpose_x = matrix_bd_57_transpose_x_0, transpose_y = matrix_bd_57_transpose_y_0, x = var_2980_cast_fp16, y = var_2982_cast_fp16)[name = string("matrix_bd_57_cast_fp16")]; - tensor matrix_bd_59_pad_0 = const()[name = string("matrix_bd_59_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_59_mode_0 = const()[name = string("matrix_bd_59_mode_0"), val = string("constant")]; - fp16 const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_59_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = matrix_bd_59_mode_0, pad = matrix_bd_59_pad_0, x = matrix_bd_57_cast_fp16)[name = string("matrix_bd_59_cast_fp16")]; - tensor var_2991 = const()[name = string("op_2991"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_61_cast_fp16 = reshape(shape = var_2991, x = matrix_bd_59_cast_fp16)[name = string("matrix_bd_61_cast_fp16")]; - tensor var_2995_begin_0 = const()[name = string("op_2995_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_2995_end_0 = const()[name = string("op_2995_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_2995_end_mask_0 = const()[name = string("op_2995_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_2995_cast_fp16 = slice_by_index(begin = var_2995_begin_0, end = var_2995_end_0, end_mask = var_2995_end_mask_0, x = matrix_bd_61_cast_fp16)[name = string("op_2995_cast_fp16")]; - tensor var_2996 = const()[name = string("op_2996"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_63_cast_fp16 = reshape(shape = var_2996, x = var_2995_cast_fp16)[name = string("matrix_bd_63_cast_fp16")]; - tensor var_3001_begin_0 = const()[name = string("op_3001_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3001_end_0 = const()[name = string("op_3001_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_3001_end_mask_0 = const()[name = string("op_3001_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_3001_cast_fp16 = slice_by_index(begin = var_3001_begin_0, end = var_3001_end_0, end_mask = var_3001_end_mask_0, x = matrix_bd_63_cast_fp16)[name = string("op_3001_cast_fp16")]; - fp16 var_3002_to_fp16 = const()[name = string("op_3002_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_15_cast_fp16 = mul(x = var_3001_cast_fp16, y = var_3002_to_fp16)[name = string("qk_mask_15_cast_fp16")]; - tensor var_3006 = const()[name = string("op_3006"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_15_cast_fp16 = reshape(shape = var_3006, x = query_31_cast_fp16)[name = string("mh_q_15_cast_fp16")]; - fp16 var_3008_to_fp16 = const()[name = string("op_3008_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_3009_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_3008_to_fp16)[name = string("op_3009_cast_fp16")]; - tensor var_3012 = const()[name = string("op_3012"), val = tensor([1, 8, 128, 188])]; - tensor var_3013_cast_fp16 = reshape(shape = var_3012, x = key_15_cast_fp16)[name = string("op_3013_cast_fp16")]; - bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; - bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; - tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_3009_cast_fp16, y = var_3013_cast_fp16)[name = string("mh_w_29_cast_fp16")]; - tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = qk_mask_15_cast_fp16)[name = string("mh_w_31_cast_fp16")]; - tensor var_3017_cast_fp16 = softmax(axis = var_2804, x = mh_w_31_cast_fp16)[name = string("op_3017_cast_fp16")]; - tensor var_3018 = const()[name = string("op_3018"), val = tensor([1, 8, 128, 188])]; - tensor var_3019_cast_fp16 = reshape(shape = var_3018, x = value_15_cast_fp16)[name = string("op_3019_cast_fp16")]; - bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; - bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; - tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_3019_cast_fp16, y = var_3017_cast_fp16)[name = string("attn_15_cast_fp16")]; - tensor var_3022 = const()[name = string("op_3022"), val = tensor([1, 1024, 1, 188])]; - tensor input_205_cast_fp16 = reshape(shape = var_3022, x = attn_15_cast_fp16)[name = string("input_205_cast_fp16")]; - string var_3032_pad_type_0 = const()[name = string("op_3032_pad_type_0"), val = string("valid")]; - tensor var_3032_strides_0 = const()[name = string("op_3032_strides_0"), val = tensor([1, 1])]; - tensor var_3032_pad_0 = const()[name = string("op_3032_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3032_dilations_0 = const()[name = string("op_3032_dilations_0"), val = tensor([1, 1])]; - int32 var_3032_groups_0 = const()[name = string("op_3032_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100738688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101131968))))[name = string("layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3032_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3032_dilations_0, groups = var_3032_groups_0, pad = var_3032_pad_0, pad_type = var_3032_pad_type_0, strides = var_3032_strides_0, weight = layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = string("op_3032_cast_fp16")]; - string var_3038_pad_type_0 = const()[name = string("op_3038_pad_type_0"), val = string("valid")]; - tensor var_3038_strides_0 = const()[name = string("op_3038_strides_0"), val = tensor([1, 1])]; - tensor var_3038_pad_0 = const()[name = string("op_3038_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3038_dilations_0 = const()[name = string("op_3038_dilations_0"), val = tensor([1, 1])]; - int32 var_3038_groups_0 = const()[name = string("op_3038_groups_0"), val = int32(1)]; - tensor layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101140608))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101133056))))[name = string("layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3038_cast_fp16 = conv(dilations = var_3038_dilations_0, groups = var_3038_groups_0, pad = var_3038_pad_0, pad_type = var_3038_pad_type_0, strides = var_3038_strides_0, weight = layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_205_cast_fp16)[name = string("op_3038_cast_fp16")]; - tensor obj_33_cast_fp16 = add(x = var_3032_cast_fp16, y = var_3038_cast_fp16)[name = string("obj_33_cast_fp16")]; - tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_33_cast_fp16)[name = string("inputs_75_cast_fp16")]; - tensor out_75_axes_0 = const()[name = string("out_75_axes_0"), val = tensor([1])]; - fp16 var_3049_to_fp16 = const()[name = string("op_3049_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_3049_to_fp16, x = inputs_75_cast_fp16)[name = string("out_75_cast_fp16")]; - tensor input_207_gamma_0_to_fp16 = const()[name = string("input_207_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101271744)))]; - tensor input_207_beta_0_to_fp16 = const()[name = string("input_207_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101273856)))]; - fp16 input_207_epsilon_0_to_fp16 = const()[name = string("input_207_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_207_cast_fp16 = batch_norm(beta = input_207_beta_0_to_fp16, epsilon = input_207_epsilon_0_to_fp16, gamma = input_207_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_75_cast_fp16)[name = string("input_207_cast_fp16")]; - string var_3070_pad_type_0 = const()[name = string("op_3070_pad_type_0"), val = string("valid")]; - tensor var_3070_strides_0 = const()[name = string("op_3070_strides_0"), val = tensor([1, 1])]; - tensor var_3070_pad_0 = const()[name = string("op_3070_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3070_dilations_0 = const()[name = string("op_3070_dilations_0"), val = tensor([1, 1])]; - int32 var_3070_groups_0 = const()[name = string("op_3070_groups_0"), val = int32(1)]; - tensor layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101275968))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102062464))))[name = string("layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3070_cast_fp16 = conv(dilations = var_3070_dilations_0, groups = var_3070_groups_0, pad = var_3070_pad_0, pad_type = var_3070_pad_type_0, strides = var_3070_strides_0, weight = layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = string("op_3070_cast_fp16")]; - string var_3076_pad_type_0 = const()[name = string("op_3076_pad_type_0"), val = string("valid")]; - tensor var_3076_strides_0 = const()[name = string("op_3076_strides_0"), val = tensor([1, 1])]; - tensor var_3076_pad_0 = const()[name = string("op_3076_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3076_dilations_0 = const()[name = string("op_3076_dilations_0"), val = tensor([1, 1])]; - int32 var_3076_groups_0 = const()[name = string("op_3076_groups_0"), val = int32(1)]; - tensor layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102083328))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102064576))))[name = string("layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3076_cast_fp16 = conv(dilations = var_3076_dilations_0, groups = var_3076_groups_0, pad = var_3076_pad_0, pad_type = var_3076_pad_type_0, strides = var_3076_strides_0, weight = layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_207_cast_fp16)[name = string("op_3076_cast_fp16")]; - tensor input_209_cast_fp16 = add(x = var_3070_cast_fp16, y = var_3076_cast_fp16)[name = string("input_209_cast_fp16")]; - int32 input_211_split_num_splits_0 = const()[name = string("input_211_split_num_splits_0"), val = int32(2)]; - int32 input_211_split_axis_0 = const()[name = string("input_211_split_axis_0"), val = int32(1)]; - tensor input_211_split_cast_fp16_0, tensor input_211_split_cast_fp16_1 = split(axis = input_211_split_axis_0, num_splits = input_211_split_num_splits_0, x = input_209_cast_fp16)[name = string("input_211_split_cast_fp16")]; - tensor input_211_split_1_sigmoid_cast_fp16 = sigmoid(x = input_211_split_cast_fp16_1)[name = string("input_211_split_1_sigmoid_cast_fp16")]; - tensor input_211_cast_fp16 = mul(x = input_211_split_cast_fp16_0, y = input_211_split_1_sigmoid_cast_fp16)[name = string("input_211_cast_fp16")]; - string input_213_pad_type_0 = const()[name = string("input_213_pad_type_0"), val = string("custom")]; - tensor input_213_pad_0 = const()[name = string("input_213_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_213_groups_0 = const()[name = string("input_213_groups_0"), val = int32(1024)]; - tensor input_213_strides_0 = const()[name = string("input_213_strides_0"), val = tensor([1, 1])]; - tensor input_213_dilations_0 = const()[name = string("input_213_dilations_0"), val = tensor([1, 1])]; - tensor const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102345536)))]; - tensor const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102364032)))]; - tensor input_215_cast_fp16 = conv(bias = const_283_to_fp16, dilations = input_213_dilations_0, groups = input_213_groups_0, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = input_213_strides_0, weight = const_282_to_fp16, x = input_211_cast_fp16)[name = string("input_215_cast_fp16")]; - tensor input_217_cast_fp16 = silu(x = input_215_cast_fp16)[name = string("input_217_cast_fp16")]; - string var_3098_pad_type_0 = const()[name = string("op_3098_pad_type_0"), val = string("valid")]; - tensor var_3098_strides_0 = const()[name = string("op_3098_strides_0"), val = tensor([1, 1])]; - tensor var_3098_pad_0 = const()[name = string("op_3098_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3098_dilations_0 = const()[name = string("op_3098_dilations_0"), val = tensor([1, 1])]; - int32 var_3098_groups_0 = const()[name = string("op_3098_groups_0"), val = int32(1)]; - tensor layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102366144))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102759424))))[name = string("layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3098_cast_fp16 = conv(dilations = var_3098_dilations_0, groups = var_3098_groups_0, pad = var_3098_pad_0, pad_type = var_3098_pad_type_0, strides = var_3098_strides_0, weight = layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = string("op_3098_cast_fp16")]; - string var_3104_pad_type_0 = const()[name = string("op_3104_pad_type_0"), val = string("valid")]; - tensor var_3104_strides_0 = const()[name = string("op_3104_strides_0"), val = tensor([1, 1])]; - tensor var_3104_pad_0 = const()[name = string("op_3104_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3104_dilations_0 = const()[name = string("op_3104_dilations_0"), val = tensor([1, 1])]; - int32 var_3104_groups_0 = const()[name = string("op_3104_groups_0"), val = int32(1)]; - tensor layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102769472))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102760512))))[name = string("layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3104_cast_fp16 = conv(dilations = var_3104_dilations_0, groups = var_3104_groups_0, pad = var_3104_pad_0, pad_type = var_3104_pad_type_0, strides = var_3104_strides_0, weight = layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_217_cast_fp16)[name = string("op_3104_cast_fp16")]; - tensor x_47_cast_fp16 = add(x = var_3098_cast_fp16, y = var_3104_cast_fp16)[name = string("x_47_cast_fp16")]; - tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = x_47_cast_fp16)[name = string("inputs_77_cast_fp16")]; - tensor out_77_axes_0 = const()[name = string("out_77_axes_0"), val = tensor([1])]; - fp16 var_3115_to_fp16 = const()[name = string("op_3115_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_3115_to_fp16, x = inputs_77_cast_fp16)[name = string("out_77_cast_fp16")]; - tensor input_219_gamma_0_to_fp16 = const()[name = string("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102900608)))]; - tensor input_219_beta_0_to_fp16 = const()[name = string("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102902720)))]; - fp16 input_219_epsilon_0_to_fp16 = const()[name = string("input_219_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_219_cast_fp16 = batch_norm(beta = input_219_beta_0_to_fp16, epsilon = input_219_epsilon_0_to_fp16, gamma = input_219_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_77_cast_fp16)[name = string("input_219_cast_fp16")]; - string var_3135_pad_type_0 = const()[name = string("op_3135_pad_type_0"), val = string("valid")]; - tensor var_3135_strides_0 = const()[name = string("op_3135_strides_0"), val = tensor([1, 1])]; - tensor var_3135_pad_0 = const()[name = string("op_3135_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3135_dilations_0 = const()[name = string("op_3135_dilations_0"), val = tensor([1, 1])]; - int32 var_3135_groups_0 = const()[name = string("op_3135_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102904832))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104477760))))[name = string("layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3135_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3135_dilations_0, groups = var_3135_groups_0, pad = var_3135_pad_0, pad_type = var_3135_pad_type_0, strides = var_3135_strides_0, weight = layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = string("op_3135_cast_fp16")]; - string var_3141_pad_type_0 = const()[name = string("op_3141_pad_type_0"), val = string("valid")]; - tensor var_3141_strides_0 = const()[name = string("op_3141_strides_0"), val = tensor([1, 1])]; - tensor var_3141_pad_0 = const()[name = string("op_3141_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3141_dilations_0 = const()[name = string("op_3141_dilations_0"), val = tensor([1, 1])]; - int32 var_3141_groups_0 = const()[name = string("op_3141_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104526912))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104481920))))[name = string("layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3141_cast_fp16 = conv(dilations = var_3141_dilations_0, groups = var_3141_groups_0, pad = var_3141_pad_0, pad_type = var_3141_pad_type_0, strides = var_3141_strides_0, weight = layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_219_cast_fp16)[name = string("op_3141_cast_fp16")]; - tensor input_221_cast_fp16 = add(x = var_3135_cast_fp16, y = var_3141_cast_fp16)[name = string("input_221_cast_fp16")]; - tensor input_223_cast_fp16 = silu(x = input_221_cast_fp16)[name = string("input_223_cast_fp16")]; - string var_3152_pad_type_0 = const()[name = string("op_3152_pad_type_0"), val = string("valid")]; - tensor var_3152_strides_0 = const()[name = string("op_3152_strides_0"), val = tensor([1, 1])]; - tensor var_3152_pad_0 = const()[name = string("op_3152_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3152_dilations_0 = const()[name = string("op_3152_dilations_0"), val = tensor([1, 1])]; - int32 var_3152_groups_0 = const()[name = string("op_3152_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105051264))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106624192))))[name = string("layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3152_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3152_dilations_0, groups = var_3152_groups_0, pad = var_3152_pad_0, pad_type = var_3152_pad_type_0, strides = var_3152_strides_0, weight = layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = string("op_3152_cast_fp16")]; - string var_3158_pad_type_0 = const()[name = string("op_3158_pad_type_0"), val = string("valid")]; - tensor var_3158_strides_0 = const()[name = string("op_3158_strides_0"), val = tensor([1, 1])]; - tensor var_3158_pad_0 = const()[name = string("op_3158_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3158_dilations_0 = const()[name = string("op_3158_dilations_0"), val = tensor([1, 1])]; - int32 var_3158_groups_0 = const()[name = string("op_3158_groups_0"), val = int32(1)]; - tensor layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106681792))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106625280))))[name = string("layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3158_cast_fp16 = conv(dilations = var_3158_dilations_0, groups = var_3158_groups_0, pad = var_3158_pad_0, pad_type = var_3158_pad_type_0, strides = var_3158_strides_0, weight = layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_223_cast_fp16)[name = string("op_3158_cast_fp16")]; - tensor x_49_cast_fp16 = add(x = var_3152_cast_fp16, y = var_3158_cast_fp16)[name = string("x_49_cast_fp16")]; - fp16 var_3160_to_fp16 = const()[name = string("op_3160_to_fp16"), val = fp16(0x1p-1)]; - tensor var_3161_cast_fp16 = mul(x = x_49_cast_fp16, y = var_3160_to_fp16)[name = string("op_3161_cast_fp16")]; - tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = var_3161_cast_fp16)[name = string("inputs_79_cast_fp16")]; - tensor out_79_axes_0 = const()[name = string("out_79_axes_0"), val = tensor([1])]; - fp16 var_3171_to_fp16 = const()[name = string("op_3171_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_3171_to_fp16, x = inputs_79_cast_fp16)[name = string("out_79_cast_fp16")]; - tensor inputs_81_gamma_0_to_fp16 = const()[name = string("inputs_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107206144)))]; - tensor inputs_81_beta_0_to_fp16 = const()[name = string("inputs_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107208256)))]; - fp16 inputs_81_epsilon_0_to_fp16 = const()[name = string("inputs_81_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_81_cast_fp16 = batch_norm(beta = inputs_81_beta_0_to_fp16, epsilon = inputs_81_epsilon_0_to_fp16, gamma = inputs_81_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_79_cast_fp16)[name = string("inputs_81_cast_fp16")]; - int32 var_3185 = const()[name = string("op_3185"), val = int32(3)]; - tensor out_81_axes_0 = const()[name = string("out_81_axes_0"), val = tensor([1])]; - fp16 var_3216_to_fp16 = const()[name = string("op_3216_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_3216_to_fp16, x = inputs_81_cast_fp16)[name = string("out_81_cast_fp16")]; - tensor input_225_gamma_0_to_fp16 = const()[name = string("input_225_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107210368)))]; - tensor input_225_beta_0_to_fp16 = const()[name = string("input_225_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107212480)))]; - fp16 input_225_epsilon_0_to_fp16 = const()[name = string("input_225_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_225_cast_fp16 = batch_norm(beta = input_225_beta_0_to_fp16, epsilon = input_225_epsilon_0_to_fp16, gamma = input_225_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_81_cast_fp16)[name = string("input_225_cast_fp16")]; - string var_3236_pad_type_0 = const()[name = string("op_3236_pad_type_0"), val = string("valid")]; - tensor var_3236_strides_0 = const()[name = string("op_3236_strides_0"), val = tensor([1, 1])]; - tensor var_3236_pad_0 = const()[name = string("op_3236_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3236_dilations_0 = const()[name = string("op_3236_dilations_0"), val = tensor([1, 1])]; - int32 var_3236_groups_0 = const()[name = string("op_3236_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107214592))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108787520))))[name = string("layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3236_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3236_dilations_0, groups = var_3236_groups_0, pad = var_3236_pad_0, pad_type = var_3236_pad_type_0, strides = var_3236_strides_0, weight = layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = string("op_3236_cast_fp16")]; - string var_3242_pad_type_0 = const()[name = string("op_3242_pad_type_0"), val = string("valid")]; - tensor var_3242_strides_0 = const()[name = string("op_3242_strides_0"), val = tensor([1, 1])]; - tensor var_3242_pad_0 = const()[name = string("op_3242_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3242_dilations_0 = const()[name = string("op_3242_dilations_0"), val = tensor([1, 1])]; - int32 var_3242_groups_0 = const()[name = string("op_3242_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108834112))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108791680))))[name = string("layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3242_cast_fp16 = conv(dilations = var_3242_dilations_0, groups = var_3242_groups_0, pad = var_3242_pad_0, pad_type = var_3242_pad_type_0, strides = var_3242_strides_0, weight = layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_225_cast_fp16)[name = string("op_3242_cast_fp16")]; - tensor input_227_cast_fp16 = add(x = var_3236_cast_fp16, y = var_3242_cast_fp16)[name = string("input_227_cast_fp16")]; - tensor input_229_cast_fp16 = silu(x = input_227_cast_fp16)[name = string("input_229_cast_fp16")]; - string var_3253_pad_type_0 = const()[name = string("op_3253_pad_type_0"), val = string("valid")]; - tensor var_3253_strides_0 = const()[name = string("op_3253_strides_0"), val = tensor([1, 1])]; - tensor var_3253_pad_0 = const()[name = string("op_3253_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3253_dilations_0 = const()[name = string("op_3253_dilations_0"), val = tensor([1, 1])]; - int32 var_3253_groups_0 = const()[name = string("op_3253_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109358464))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110931392))))[name = string("layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3253_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3253_dilations_0, groups = var_3253_groups_0, pad = var_3253_pad_0, pad_type = var_3253_pad_type_0, strides = var_3253_strides_0, weight = layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = string("op_3253_cast_fp16")]; - string var_3259_pad_type_0 = const()[name = string("op_3259_pad_type_0"), val = string("valid")]; - tensor var_3259_strides_0 = const()[name = string("op_3259_strides_0"), val = tensor([1, 1])]; - tensor var_3259_pad_0 = const()[name = string("op_3259_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3259_dilations_0 = const()[name = string("op_3259_dilations_0"), val = tensor([1, 1])]; - int32 var_3259_groups_0 = const()[name = string("op_3259_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111002944))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110932480))))[name = string("layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3259_cast_fp16 = conv(dilations = var_3259_dilations_0, groups = var_3259_groups_0, pad = var_3259_pad_0, pad_type = var_3259_pad_type_0, strides = var_3259_strides_0, weight = layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_229_cast_fp16)[name = string("op_3259_cast_fp16")]; - tensor x_51_cast_fp16 = add(x = var_3253_cast_fp16, y = var_3259_cast_fp16)[name = string("x_51_cast_fp16")]; - fp16 var_3261_to_fp16 = const()[name = string("op_3261_to_fp16"), val = fp16(0x1p-1)]; - tensor var_3262_cast_fp16 = mul(x = x_51_cast_fp16, y = var_3261_to_fp16)[name = string("op_3262_cast_fp16")]; - tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = var_3262_cast_fp16)[name = string("inputs_83_cast_fp16")]; - tensor out_83_axes_0 = const()[name = string("out_83_axes_0"), val = tensor([1])]; - fp16 var_3272_to_fp16 = const()[name = string("op_3272_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_3272_to_fp16, x = inputs_83_cast_fp16)[name = string("out_83_cast_fp16")]; - tensor obj_35_gamma_0_to_fp16 = const()[name = string("obj_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111527296)))]; - tensor obj_35_beta_0_to_fp16 = const()[name = string("obj_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111529408)))]; - fp16 obj_35_epsilon_0_to_fp16 = const()[name = string("obj_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_35_cast_fp16 = batch_norm(beta = obj_35_beta_0_to_fp16, epsilon = obj_35_epsilon_0_to_fp16, gamma = obj_35_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_83_cast_fp16)[name = string("obj_35_cast_fp16")]; - string var_3297_pad_type_0 = const()[name = string("op_3297_pad_type_0"), val = string("valid")]; - tensor var_3297_strides_0 = const()[name = string("op_3297_strides_0"), val = tensor([1, 1])]; - tensor var_3297_pad_0 = const()[name = string("op_3297_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3297_dilations_0 = const()[name = string("op_3297_dilations_0"), val = tensor([1, 1])]; - int32 var_3297_groups_0 = const()[name = string("op_3297_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111531520))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111924800))))[name = string("layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3297_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3297_dilations_0, groups = var_3297_groups_0, pad = var_3297_pad_0, pad_type = var_3297_pad_type_0, strides = var_3297_strides_0, weight = layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = string("op_3297_cast_fp16")]; - string var_3303_pad_type_0 = const()[name = string("op_3303_pad_type_0"), val = string("valid")]; - tensor var_3303_strides_0 = const()[name = string("op_3303_strides_0"), val = tensor([1, 1])]; - tensor var_3303_pad_0 = const()[name = string("op_3303_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3303_dilations_0 = const()[name = string("op_3303_dilations_0"), val = tensor([1, 1])]; - int32 var_3303_groups_0 = const()[name = string("op_3303_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111939712))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111925888))))[name = string("layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3303_cast_fp16 = conv(dilations = var_3303_dilations_0, groups = var_3303_groups_0, pad = var_3303_pad_0, pad_type = var_3303_pad_type_0, strides = var_3303_strides_0, weight = layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = string("op_3303_cast_fp16")]; - tensor query_33_cast_fp16 = add(x = var_3297_cast_fp16, y = var_3303_cast_fp16)[name = string("query_33_cast_fp16")]; - string var_3312_pad_type_0 = const()[name = string("op_3312_pad_type_0"), val = string("valid")]; - tensor var_3312_strides_0 = const()[name = string("op_3312_strides_0"), val = tensor([1, 1])]; - tensor var_3312_pad_0 = const()[name = string("op_3312_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3312_dilations_0 = const()[name = string("op_3312_dilations_0"), val = tensor([1, 1])]; - int32 var_3312_groups_0 = const()[name = string("op_3312_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112070848))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112464128))))[name = string("layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3312_cast_fp16 = conv(dilations = var_3312_dilations_0, groups = var_3312_groups_0, pad = var_3312_pad_0, pad_type = var_3312_pad_type_0, strides = var_3312_strides_0, weight = layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = string("op_3312_cast_fp16")]; - string var_3318_pad_type_0 = const()[name = string("op_3318_pad_type_0"), val = string("valid")]; - tensor var_3318_strides_0 = const()[name = string("op_3318_strides_0"), val = tensor([1, 1])]; - tensor var_3318_pad_0 = const()[name = string("op_3318_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3318_dilations_0 = const()[name = string("op_3318_dilations_0"), val = tensor([1, 1])]; - int32 var_3318_groups_0 = const()[name = string("op_3318_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112485440))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112465216))))[name = string("layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3318_cast_fp16 = conv(dilations = var_3318_dilations_0, groups = var_3318_groups_0, pad = var_3318_pad_0, pad_type = var_3318_pad_type_0, strides = var_3318_strides_0, weight = layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = string("op_3318_cast_fp16")]; - tensor key_17_cast_fp16 = add(x = var_3312_cast_fp16, y = var_3318_cast_fp16)[name = string("key_17_cast_fp16")]; - string var_3328_pad_type_0 = const()[name = string("op_3328_pad_type_0"), val = string("valid")]; - tensor var_3328_strides_0 = const()[name = string("op_3328_strides_0"), val = tensor([1, 1])]; - tensor var_3328_pad_0 = const()[name = string("op_3328_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3328_dilations_0 = const()[name = string("op_3328_dilations_0"), val = tensor([1, 1])]; - int32 var_3328_groups_0 = const()[name = string("op_3328_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112616576))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113009856))))[name = string("layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3328_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3328_dilations_0, groups = var_3328_groups_0, pad = var_3328_pad_0, pad_type = var_3328_pad_type_0, strides = var_3328_strides_0, weight = layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = string("op_3328_cast_fp16")]; - string var_3334_pad_type_0 = const()[name = string("op_3334_pad_type_0"), val = string("valid")]; - tensor var_3334_strides_0 = const()[name = string("op_3334_strides_0"), val = tensor([1, 1])]; - tensor var_3334_pad_0 = const()[name = string("op_3334_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3334_dilations_0 = const()[name = string("op_3334_dilations_0"), val = tensor([1, 1])]; - int32 var_3334_groups_0 = const()[name = string("op_3334_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113022656))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113010944))))[name = string("layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3334_cast_fp16 = conv(dilations = var_3334_dilations_0, groups = var_3334_groups_0, pad = var_3334_pad_0, pad_type = var_3334_pad_type_0, strides = var_3334_strides_0, weight = layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = string("op_3334_cast_fp16")]; - tensor value_17_cast_fp16 = add(x = var_3328_cast_fp16, y = var_3334_cast_fp16)[name = string("value_17_cast_fp16")]; - tensor var_3337_to_fp16 = const()[name = string("op_3337_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113153792)))]; - tensor query_35_cast_fp16 = add(x = query_33_cast_fp16, y = var_3337_to_fp16)[name = string("query_35_cast_fp16")]; - tensor var_3340_to_fp16 = const()[name = string("op_3340_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113155904)))]; - tensor q_with_bias_v_17_cast_fp16 = add(x = query_33_cast_fp16, y = var_3340_to_fp16)[name = string("q_with_bias_v_17_cast_fp16")]; - string var_3350_pad_type_0 = const()[name = string("op_3350_pad_type_0"), val = string("valid")]; - tensor var_3350_strides_0 = const()[name = string("op_3350_strides_0"), val = tensor([1, 1])]; - tensor var_3350_pad_0 = const()[name = string("op_3350_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3350_dilations_0 = const()[name = string("op_3350_dilations_0"), val = tensor([1, 1])]; - int32 var_3350_groups_0 = const()[name = string("op_3350_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113158016))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113551296))))[name = string("layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_3350_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3350_dilations_0, groups = var_3350_groups_0, pad = var_3350_pad_0, pad_type = var_3350_pad_type_0, strides = var_3350_strides_0, weight = layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_3350_cast_fp16")]; - string var_3356_pad_type_0 = const()[name = string("op_3356_pad_type_0"), val = string("valid")]; - tensor var_3356_strides_0 = const()[name = string("op_3356_strides_0"), val = tensor([1, 1])]; - tensor var_3356_pad_0 = const()[name = string("op_3356_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3356_dilations_0 = const()[name = string("op_3356_dilations_0"), val = tensor([1, 1])]; - int32 var_3356_groups_0 = const()[name = string("op_3356_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113589632))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113552384))))[name = string("layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3356_cast_fp16 = conv(dilations = var_3356_dilations_0, groups = var_3356_groups_0, pad = var_3356_pad_0, pad_type = var_3356_pad_type_0, strides = var_3356_strides_0, weight = layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_3356_cast_fp16")]; - tensor p_17_cast_fp16 = add(x = var_3350_cast_fp16, y = var_3356_cast_fp16)[name = string("p_17_cast_fp16")]; - tensor var_3360 = const()[name = string("op_3360"), val = tensor([1, 8, 128, 188])]; - tensor var_3361_cast_fp16 = reshape(shape = var_3360, x = q_with_bias_v_17_cast_fp16)[name = string("op_3361_cast_fp16")]; - tensor var_3362 = const()[name = string("op_3362"), val = tensor([1, 8, 128, -1])]; - tensor var_3363_cast_fp16 = reshape(shape = var_3362, x = p_17_cast_fp16)[name = string("op_3363_cast_fp16")]; - bool matrix_bd_65_transpose_x_0 = const()[name = string("matrix_bd_65_transpose_x_0"), val = bool(true)]; - bool matrix_bd_65_transpose_y_0 = const()[name = string("matrix_bd_65_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_65_cast_fp16 = matmul(transpose_x = matrix_bd_65_transpose_x_0, transpose_y = matrix_bd_65_transpose_y_0, x = var_3361_cast_fp16, y = var_3363_cast_fp16)[name = string("matrix_bd_65_cast_fp16")]; - tensor matrix_bd_67_pad_0 = const()[name = string("matrix_bd_67_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_67_mode_0 = const()[name = string("matrix_bd_67_mode_0"), val = string("constant")]; - fp16 const_98_to_fp16 = const()[name = string("const_98_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_67_cast_fp16 = pad(constant_val = const_98_to_fp16, mode = matrix_bd_67_mode_0, pad = matrix_bd_67_pad_0, x = matrix_bd_65_cast_fp16)[name = string("matrix_bd_67_cast_fp16")]; - tensor var_3372 = const()[name = string("op_3372"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_69_cast_fp16 = reshape(shape = var_3372, x = matrix_bd_67_cast_fp16)[name = string("matrix_bd_69_cast_fp16")]; - tensor var_3376_begin_0 = const()[name = string("op_3376_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_3376_end_0 = const()[name = string("op_3376_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_3376_end_mask_0 = const()[name = string("op_3376_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_3376_cast_fp16 = slice_by_index(begin = var_3376_begin_0, end = var_3376_end_0, end_mask = var_3376_end_mask_0, x = matrix_bd_69_cast_fp16)[name = string("op_3376_cast_fp16")]; - tensor var_3377 = const()[name = string("op_3377"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_71_cast_fp16 = reshape(shape = var_3377, x = var_3376_cast_fp16)[name = string("matrix_bd_71_cast_fp16")]; - tensor var_3382_begin_0 = const()[name = string("op_3382_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3382_end_0 = const()[name = string("op_3382_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_3382_end_mask_0 = const()[name = string("op_3382_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_3382_cast_fp16 = slice_by_index(begin = var_3382_begin_0, end = var_3382_end_0, end_mask = var_3382_end_mask_0, x = matrix_bd_71_cast_fp16)[name = string("op_3382_cast_fp16")]; - fp16 var_3383_to_fp16 = const()[name = string("op_3383_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_17_cast_fp16 = mul(x = var_3382_cast_fp16, y = var_3383_to_fp16)[name = string("qk_mask_17_cast_fp16")]; - tensor var_3387 = const()[name = string("op_3387"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_17_cast_fp16 = reshape(shape = var_3387, x = query_35_cast_fp16)[name = string("mh_q_17_cast_fp16")]; - fp16 var_3389_to_fp16 = const()[name = string("op_3389_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_3390_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_3389_to_fp16)[name = string("op_3390_cast_fp16")]; - tensor var_3393 = const()[name = string("op_3393"), val = tensor([1, 8, 128, 188])]; - tensor var_3394_cast_fp16 = reshape(shape = var_3393, x = key_17_cast_fp16)[name = string("op_3394_cast_fp16")]; - bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; - bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; - tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_3390_cast_fp16, y = var_3394_cast_fp16)[name = string("mh_w_33_cast_fp16")]; - tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = qk_mask_17_cast_fp16)[name = string("mh_w_35_cast_fp16")]; - tensor var_3398_cast_fp16 = softmax(axis = var_3185, x = mh_w_35_cast_fp16)[name = string("op_3398_cast_fp16")]; - tensor var_3399 = const()[name = string("op_3399"), val = tensor([1, 8, 128, 188])]; - tensor var_3400_cast_fp16 = reshape(shape = var_3399, x = value_17_cast_fp16)[name = string("op_3400_cast_fp16")]; - bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; - bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; - tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_3400_cast_fp16, y = var_3398_cast_fp16)[name = string("attn_17_cast_fp16")]; - tensor var_3403 = const()[name = string("op_3403"), val = tensor([1, 1024, 1, 188])]; - tensor input_231_cast_fp16 = reshape(shape = var_3403, x = attn_17_cast_fp16)[name = string("input_231_cast_fp16")]; - string var_3413_pad_type_0 = const()[name = string("op_3413_pad_type_0"), val = string("valid")]; - tensor var_3413_strides_0 = const()[name = string("op_3413_strides_0"), val = tensor([1, 1])]; - tensor var_3413_pad_0 = const()[name = string("op_3413_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3413_dilations_0 = const()[name = string("op_3413_dilations_0"), val = tensor([1, 1])]; - int32 var_3413_groups_0 = const()[name = string("op_3413_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113720768))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114114048))))[name = string("layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3413_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3413_dilations_0, groups = var_3413_groups_0, pad = var_3413_pad_0, pad_type = var_3413_pad_type_0, strides = var_3413_strides_0, weight = layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_231_cast_fp16)[name = string("op_3413_cast_fp16")]; - string var_3419_pad_type_0 = const()[name = string("op_3419_pad_type_0"), val = string("valid")]; - tensor var_3419_strides_0 = const()[name = string("op_3419_strides_0"), val = tensor([1, 1])]; - tensor var_3419_pad_0 = const()[name = string("op_3419_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3419_dilations_0 = const()[name = string("op_3419_dilations_0"), val = tensor([1, 1])]; - int32 var_3419_groups_0 = const()[name = string("op_3419_groups_0"), val = int32(1)]; - tensor layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114123968))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114115136))))[name = string("layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3419_cast_fp16 = conv(dilations = var_3419_dilations_0, groups = var_3419_groups_0, pad = var_3419_pad_0, pad_type = var_3419_pad_type_0, strides = var_3419_strides_0, weight = layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_231_cast_fp16)[name = string("op_3419_cast_fp16")]; - tensor obj_37_cast_fp16 = add(x = var_3413_cast_fp16, y = var_3419_cast_fp16)[name = string("obj_37_cast_fp16")]; - tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = obj_37_cast_fp16)[name = string("inputs_85_cast_fp16")]; - tensor out_85_axes_0 = const()[name = string("out_85_axes_0"), val = tensor([1])]; - fp16 var_3430_to_fp16 = const()[name = string("op_3430_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_3430_to_fp16, x = inputs_85_cast_fp16)[name = string("out_85_cast_fp16")]; - tensor input_233_gamma_0_to_fp16 = const()[name = string("input_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114255104)))]; - tensor input_233_beta_0_to_fp16 = const()[name = string("input_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114257216)))]; - fp16 input_233_epsilon_0_to_fp16 = const()[name = string("input_233_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_233_cast_fp16 = batch_norm(beta = input_233_beta_0_to_fp16, epsilon = input_233_epsilon_0_to_fp16, gamma = input_233_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_85_cast_fp16)[name = string("input_233_cast_fp16")]; - string var_3451_pad_type_0 = const()[name = string("op_3451_pad_type_0"), val = string("valid")]; - tensor var_3451_strides_0 = const()[name = string("op_3451_strides_0"), val = tensor([1, 1])]; - tensor var_3451_pad_0 = const()[name = string("op_3451_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3451_dilations_0 = const()[name = string("op_3451_dilations_0"), val = tensor([1, 1])]; - int32 var_3451_groups_0 = const()[name = string("op_3451_groups_0"), val = int32(1)]; - tensor layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114259328))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115045824))))[name = string("layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3451_cast_fp16 = conv(dilations = var_3451_dilations_0, groups = var_3451_groups_0, pad = var_3451_pad_0, pad_type = var_3451_pad_type_0, strides = var_3451_strides_0, weight = layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = string("op_3451_cast_fp16")]; - string var_3457_pad_type_0 = const()[name = string("op_3457_pad_type_0"), val = string("valid")]; - tensor var_3457_strides_0 = const()[name = string("op_3457_strides_0"), val = tensor([1, 1])]; - tensor var_3457_pad_0 = const()[name = string("op_3457_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3457_dilations_0 = const()[name = string("op_3457_dilations_0"), val = tensor([1, 1])]; - int32 var_3457_groups_0 = const()[name = string("op_3457_groups_0"), val = int32(1)]; - tensor layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115064192))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115047936))))[name = string("layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3457_cast_fp16 = conv(dilations = var_3457_dilations_0, groups = var_3457_groups_0, pad = var_3457_pad_0, pad_type = var_3457_pad_type_0, strides = var_3457_strides_0, weight = layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_233_cast_fp16)[name = string("op_3457_cast_fp16")]; - tensor input_235_cast_fp16 = add(x = var_3451_cast_fp16, y = var_3457_cast_fp16)[name = string("input_235_cast_fp16")]; - int32 input_237_split_num_splits_0 = const()[name = string("input_237_split_num_splits_0"), val = int32(2)]; - int32 input_237_split_axis_0 = const()[name = string("input_237_split_axis_0"), val = int32(1)]; - tensor input_237_split_cast_fp16_0, tensor input_237_split_cast_fp16_1 = split(axis = input_237_split_axis_0, num_splits = input_237_split_num_splits_0, x = input_235_cast_fp16)[name = string("input_237_split_cast_fp16")]; - tensor input_237_split_1_sigmoid_cast_fp16 = sigmoid(x = input_237_split_cast_fp16_1)[name = string("input_237_split_1_sigmoid_cast_fp16")]; - tensor input_237_cast_fp16 = mul(x = input_237_split_cast_fp16_0, y = input_237_split_1_sigmoid_cast_fp16)[name = string("input_237_cast_fp16")]; - string input_239_pad_type_0 = const()[name = string("input_239_pad_type_0"), val = string("custom")]; - tensor input_239_pad_0 = const()[name = string("input_239_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_239_groups_0 = const()[name = string("input_239_groups_0"), val = int32(1024)]; - tensor input_239_strides_0 = const()[name = string("input_239_strides_0"), val = tensor([1, 1])]; - tensor input_239_dilations_0 = const()[name = string("input_239_dilations_0"), val = tensor([1, 1])]; - tensor const_284_to_fp16 = const()[name = string("const_284_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115326400)))]; - tensor const_285_to_fp16 = const()[name = string("const_285_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115344896)))]; - tensor input_241_cast_fp16 = conv(bias = const_285_to_fp16, dilations = input_239_dilations_0, groups = input_239_groups_0, pad = input_239_pad_0, pad_type = input_239_pad_type_0, strides = input_239_strides_0, weight = const_284_to_fp16, x = input_237_cast_fp16)[name = string("input_241_cast_fp16")]; - tensor input_243_cast_fp16 = silu(x = input_241_cast_fp16)[name = string("input_243_cast_fp16")]; - string var_3479_pad_type_0 = const()[name = string("op_3479_pad_type_0"), val = string("valid")]; - tensor var_3479_strides_0 = const()[name = string("op_3479_strides_0"), val = tensor([1, 1])]; - tensor var_3479_pad_0 = const()[name = string("op_3479_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3479_dilations_0 = const()[name = string("op_3479_dilations_0"), val = tensor([1, 1])]; - int32 var_3479_groups_0 = const()[name = string("op_3479_groups_0"), val = int32(1)]; - tensor layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115347008))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115740288))))[name = string("layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3479_cast_fp16 = conv(dilations = var_3479_dilations_0, groups = var_3479_groups_0, pad = var_3479_pad_0, pad_type = var_3479_pad_type_0, strides = var_3479_strides_0, weight = layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_243_cast_fp16)[name = string("op_3479_cast_fp16")]; - string var_3485_pad_type_0 = const()[name = string("op_3485_pad_type_0"), val = string("valid")]; - tensor var_3485_strides_0 = const()[name = string("op_3485_strides_0"), val = tensor([1, 1])]; - tensor var_3485_pad_0 = const()[name = string("op_3485_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3485_dilations_0 = const()[name = string("op_3485_dilations_0"), val = tensor([1, 1])]; - int32 var_3485_groups_0 = const()[name = string("op_3485_groups_0"), val = int32(1)]; - tensor layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115749632))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115741376))))[name = string("layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3485_cast_fp16 = conv(dilations = var_3485_dilations_0, groups = var_3485_groups_0, pad = var_3485_pad_0, pad_type = var_3485_pad_type_0, strides = var_3485_strides_0, weight = layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_243_cast_fp16)[name = string("op_3485_cast_fp16")]; - tensor x_53_cast_fp16 = add(x = var_3479_cast_fp16, y = var_3485_cast_fp16)[name = string("x_53_cast_fp16")]; - tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = x_53_cast_fp16)[name = string("inputs_87_cast_fp16")]; - tensor out_87_axes_0 = const()[name = string("out_87_axes_0"), val = tensor([1])]; - fp16 var_3496_to_fp16 = const()[name = string("op_3496_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_3496_to_fp16, x = inputs_87_cast_fp16)[name = string("out_87_cast_fp16")]; - tensor input_245_gamma_0_to_fp16 = const()[name = string("input_245_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115880768)))]; - tensor input_245_beta_0_to_fp16 = const()[name = string("input_245_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115882880)))]; - fp16 input_245_epsilon_0_to_fp16 = const()[name = string("input_245_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_245_cast_fp16 = batch_norm(beta = input_245_beta_0_to_fp16, epsilon = input_245_epsilon_0_to_fp16, gamma = input_245_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_87_cast_fp16)[name = string("input_245_cast_fp16")]; - string var_3516_pad_type_0 = const()[name = string("op_3516_pad_type_0"), val = string("valid")]; - tensor var_3516_strides_0 = const()[name = string("op_3516_strides_0"), val = tensor([1, 1])]; - tensor var_3516_pad_0 = const()[name = string("op_3516_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3516_dilations_0 = const()[name = string("op_3516_dilations_0"), val = tensor([1, 1])]; - int32 var_3516_groups_0 = const()[name = string("op_3516_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115884992))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117457920))))[name = string("layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3516_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3516_dilations_0, groups = var_3516_groups_0, pad = var_3516_pad_0, pad_type = var_3516_pad_type_0, strides = var_3516_strides_0, weight = layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_245_cast_fp16)[name = string("op_3516_cast_fp16")]; - string var_3522_pad_type_0 = const()[name = string("op_3522_pad_type_0"), val = string("valid")]; - tensor var_3522_strides_0 = const()[name = string("op_3522_strides_0"), val = tensor([1, 1])]; - tensor var_3522_pad_0 = const()[name = string("op_3522_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3522_dilations_0 = const()[name = string("op_3522_dilations_0"), val = tensor([1, 1])]; - int32 var_3522_groups_0 = const()[name = string("op_3522_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117507264))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117462080))))[name = string("layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3522_cast_fp16 = conv(dilations = var_3522_dilations_0, groups = var_3522_groups_0, pad = var_3522_pad_0, pad_type = var_3522_pad_type_0, strides = var_3522_strides_0, weight = layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_245_cast_fp16)[name = string("op_3522_cast_fp16")]; - tensor input_247_cast_fp16 = add(x = var_3516_cast_fp16, y = var_3522_cast_fp16)[name = string("input_247_cast_fp16")]; - tensor input_249_cast_fp16 = silu(x = input_247_cast_fp16)[name = string("input_249_cast_fp16")]; - string var_3533_pad_type_0 = const()[name = string("op_3533_pad_type_0"), val = string("valid")]; - tensor var_3533_strides_0 = const()[name = string("op_3533_strides_0"), val = tensor([1, 1])]; - tensor var_3533_pad_0 = const()[name = string("op_3533_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3533_dilations_0 = const()[name = string("op_3533_dilations_0"), val = tensor([1, 1])]; - int32 var_3533_groups_0 = const()[name = string("op_3533_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118031616))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119604544))))[name = string("layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3533_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3533_dilations_0, groups = var_3533_groups_0, pad = var_3533_pad_0, pad_type = var_3533_pad_type_0, strides = var_3533_strides_0, weight = layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = string("op_3533_cast_fp16")]; - string var_3539_pad_type_0 = const()[name = string("op_3539_pad_type_0"), val = string("valid")]; - tensor var_3539_strides_0 = const()[name = string("op_3539_strides_0"), val = tensor([1, 1])]; - tensor var_3539_pad_0 = const()[name = string("op_3539_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3539_dilations_0 = const()[name = string("op_3539_dilations_0"), val = tensor([1, 1])]; - int32 var_3539_groups_0 = const()[name = string("op_3539_groups_0"), val = int32(1)]; - tensor layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119656000))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119605632))))[name = string("layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3539_cast_fp16 = conv(dilations = var_3539_dilations_0, groups = var_3539_groups_0, pad = var_3539_pad_0, pad_type = var_3539_pad_type_0, strides = var_3539_strides_0, weight = layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_249_cast_fp16)[name = string("op_3539_cast_fp16")]; - tensor x_55_cast_fp16 = add(x = var_3533_cast_fp16, y = var_3539_cast_fp16)[name = string("x_55_cast_fp16")]; - fp16 var_3541_to_fp16 = const()[name = string("op_3541_to_fp16"), val = fp16(0x1p-1)]; - tensor var_3542_cast_fp16 = mul(x = x_55_cast_fp16, y = var_3541_to_fp16)[name = string("op_3542_cast_fp16")]; - tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = var_3542_cast_fp16)[name = string("inputs_89_cast_fp16")]; - tensor out_89_axes_0 = const()[name = string("out_89_axes_0"), val = tensor([1])]; - fp16 var_3552_to_fp16 = const()[name = string("op_3552_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_3552_to_fp16, x = inputs_89_cast_fp16)[name = string("out_89_cast_fp16")]; - tensor inputs_91_gamma_0_to_fp16 = const()[name = string("inputs_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120180352)))]; - tensor inputs_91_beta_0_to_fp16 = const()[name = string("inputs_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120182464)))]; - fp16 inputs_91_epsilon_0_to_fp16 = const()[name = string("inputs_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_91_cast_fp16 = batch_norm(beta = inputs_91_beta_0_to_fp16, epsilon = inputs_91_epsilon_0_to_fp16, gamma = inputs_91_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_89_cast_fp16)[name = string("inputs_91_cast_fp16")]; - int32 var_3566 = const()[name = string("op_3566"), val = int32(3)]; - tensor out_91_axes_0 = const()[name = string("out_91_axes_0"), val = tensor([1])]; - fp16 var_3597_to_fp16 = const()[name = string("op_3597_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_3597_to_fp16, x = inputs_91_cast_fp16)[name = string("out_91_cast_fp16")]; - tensor input_251_gamma_0_to_fp16 = const()[name = string("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120184576)))]; - tensor input_251_beta_0_to_fp16 = const()[name = string("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120186688)))]; - fp16 input_251_epsilon_0_to_fp16 = const()[name = string("input_251_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_91_cast_fp16)[name = string("input_251_cast_fp16")]; - string var_3617_pad_type_0 = const()[name = string("op_3617_pad_type_0"), val = string("valid")]; - tensor var_3617_strides_0 = const()[name = string("op_3617_strides_0"), val = tensor([1, 1])]; - tensor var_3617_pad_0 = const()[name = string("op_3617_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3617_dilations_0 = const()[name = string("op_3617_dilations_0"), val = tensor([1, 1])]; - int32 var_3617_groups_0 = const()[name = string("op_3617_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120188800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121761728))))[name = string("layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3617_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3617_dilations_0, groups = var_3617_groups_0, pad = var_3617_pad_0, pad_type = var_3617_pad_type_0, strides = var_3617_strides_0, weight = layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_251_cast_fp16)[name = string("op_3617_cast_fp16")]; - string var_3623_pad_type_0 = const()[name = string("op_3623_pad_type_0"), val = string("valid")]; - tensor var_3623_strides_0 = const()[name = string("op_3623_strides_0"), val = tensor([1, 1])]; - tensor var_3623_pad_0 = const()[name = string("op_3623_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3623_dilations_0 = const()[name = string("op_3623_dilations_0"), val = tensor([1, 1])]; - int32 var_3623_groups_0 = const()[name = string("op_3623_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121813184))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121765888))))[name = string("layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3623_cast_fp16 = conv(dilations = var_3623_dilations_0, groups = var_3623_groups_0, pad = var_3623_pad_0, pad_type = var_3623_pad_type_0, strides = var_3623_strides_0, weight = layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_251_cast_fp16)[name = string("op_3623_cast_fp16")]; - tensor input_253_cast_fp16 = add(x = var_3617_cast_fp16, y = var_3623_cast_fp16)[name = string("input_253_cast_fp16")]; - tensor input_255_cast_fp16 = silu(x = input_253_cast_fp16)[name = string("input_255_cast_fp16")]; - string var_3634_pad_type_0 = const()[name = string("op_3634_pad_type_0"), val = string("valid")]; - tensor var_3634_strides_0 = const()[name = string("op_3634_strides_0"), val = tensor([1, 1])]; - tensor var_3634_pad_0 = const()[name = string("op_3634_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3634_dilations_0 = const()[name = string("op_3634_dilations_0"), val = tensor([1, 1])]; - int32 var_3634_groups_0 = const()[name = string("op_3634_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122337536))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123910464))))[name = string("layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3634_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3634_dilations_0, groups = var_3634_groups_0, pad = var_3634_pad_0, pad_type = var_3634_pad_type_0, strides = var_3634_strides_0, weight = layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_255_cast_fp16)[name = string("op_3634_cast_fp16")]; - string var_3640_pad_type_0 = const()[name = string("op_3640_pad_type_0"), val = string("valid")]; - tensor var_3640_strides_0 = const()[name = string("op_3640_strides_0"), val = tensor([1, 1])]; - tensor var_3640_pad_0 = const()[name = string("op_3640_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3640_dilations_0 = const()[name = string("op_3640_dilations_0"), val = tensor([1, 1])]; - int32 var_3640_groups_0 = const()[name = string("op_3640_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123975616))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123911552))))[name = string("layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3640_cast_fp16 = conv(dilations = var_3640_dilations_0, groups = var_3640_groups_0, pad = var_3640_pad_0, pad_type = var_3640_pad_type_0, strides = var_3640_strides_0, weight = layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_255_cast_fp16)[name = string("op_3640_cast_fp16")]; - tensor x_57_cast_fp16 = add(x = var_3634_cast_fp16, y = var_3640_cast_fp16)[name = string("x_57_cast_fp16")]; - fp16 var_3642_to_fp16 = const()[name = string("op_3642_to_fp16"), val = fp16(0x1p-1)]; - tensor var_3643_cast_fp16 = mul(x = x_57_cast_fp16, y = var_3642_to_fp16)[name = string("op_3643_cast_fp16")]; - tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = var_3643_cast_fp16)[name = string("inputs_93_cast_fp16")]; - tensor out_93_axes_0 = const()[name = string("out_93_axes_0"), val = tensor([1])]; - fp16 var_3653_to_fp16 = const()[name = string("op_3653_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_3653_to_fp16, x = inputs_93_cast_fp16)[name = string("out_93_cast_fp16")]; - tensor obj_39_gamma_0_to_fp16 = const()[name = string("obj_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124499968)))]; - tensor obj_39_beta_0_to_fp16 = const()[name = string("obj_39_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124502080)))]; - fp16 obj_39_epsilon_0_to_fp16 = const()[name = string("obj_39_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_39_cast_fp16 = batch_norm(beta = obj_39_beta_0_to_fp16, epsilon = obj_39_epsilon_0_to_fp16, gamma = obj_39_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_93_cast_fp16)[name = string("obj_39_cast_fp16")]; - string var_3678_pad_type_0 = const()[name = string("op_3678_pad_type_0"), val = string("valid")]; - tensor var_3678_strides_0 = const()[name = string("op_3678_strides_0"), val = tensor([1, 1])]; - tensor var_3678_pad_0 = const()[name = string("op_3678_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3678_dilations_0 = const()[name = string("op_3678_dilations_0"), val = tensor([1, 1])]; - int32 var_3678_groups_0 = const()[name = string("op_3678_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124504192))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124897472))))[name = string("layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3678_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3678_dilations_0, groups = var_3678_groups_0, pad = var_3678_pad_0, pad_type = var_3678_pad_type_0, strides = var_3678_strides_0, weight = layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = string("op_3678_cast_fp16")]; - string var_3684_pad_type_0 = const()[name = string("op_3684_pad_type_0"), val = string("valid")]; - tensor var_3684_strides_0 = const()[name = string("op_3684_strides_0"), val = tensor([1, 1])]; - tensor var_3684_pad_0 = const()[name = string("op_3684_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3684_dilations_0 = const()[name = string("op_3684_dilations_0"), val = tensor([1, 1])]; - int32 var_3684_groups_0 = const()[name = string("op_3684_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124918208))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124898560))))[name = string("layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3684_cast_fp16 = conv(dilations = var_3684_dilations_0, groups = var_3684_groups_0, pad = var_3684_pad_0, pad_type = var_3684_pad_type_0, strides = var_3684_strides_0, weight = layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = string("op_3684_cast_fp16")]; - tensor query_37_cast_fp16 = add(x = var_3678_cast_fp16, y = var_3684_cast_fp16)[name = string("query_37_cast_fp16")]; - string var_3693_pad_type_0 = const()[name = string("op_3693_pad_type_0"), val = string("valid")]; - tensor var_3693_strides_0 = const()[name = string("op_3693_strides_0"), val = tensor([1, 1])]; - tensor var_3693_pad_0 = const()[name = string("op_3693_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3693_dilations_0 = const()[name = string("op_3693_dilations_0"), val = tensor([1, 1])]; - int32 var_3693_groups_0 = const()[name = string("op_3693_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125049344))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125442624))))[name = string("layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3693_cast_fp16 = conv(dilations = var_3693_dilations_0, groups = var_3693_groups_0, pad = var_3693_pad_0, pad_type = var_3693_pad_type_0, strides = var_3693_strides_0, weight = layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = string("op_3693_cast_fp16")]; - string var_3699_pad_type_0 = const()[name = string("op_3699_pad_type_0"), val = string("valid")]; - tensor var_3699_strides_0 = const()[name = string("op_3699_strides_0"), val = tensor([1, 1])]; - tensor var_3699_pad_0 = const()[name = string("op_3699_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3699_dilations_0 = const()[name = string("op_3699_dilations_0"), val = tensor([1, 1])]; - int32 var_3699_groups_0 = const()[name = string("op_3699_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125474816))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125443712))))[name = string("layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3699_cast_fp16 = conv(dilations = var_3699_dilations_0, groups = var_3699_groups_0, pad = var_3699_pad_0, pad_type = var_3699_pad_type_0, strides = var_3699_strides_0, weight = layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = string("op_3699_cast_fp16")]; - tensor key_19_cast_fp16 = add(x = var_3693_cast_fp16, y = var_3699_cast_fp16)[name = string("key_19_cast_fp16")]; - string var_3709_pad_type_0 = const()[name = string("op_3709_pad_type_0"), val = string("valid")]; - tensor var_3709_strides_0 = const()[name = string("op_3709_strides_0"), val = tensor([1, 1])]; - tensor var_3709_pad_0 = const()[name = string("op_3709_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3709_dilations_0 = const()[name = string("op_3709_dilations_0"), val = tensor([1, 1])]; - int32 var_3709_groups_0 = const()[name = string("op_3709_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125605952))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125999232))))[name = string("layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3709_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3709_dilations_0, groups = var_3709_groups_0, pad = var_3709_pad_0, pad_type = var_3709_pad_type_0, strides = var_3709_strides_0, weight = layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = string("op_3709_cast_fp16")]; - string var_3715_pad_type_0 = const()[name = string("op_3715_pad_type_0"), val = string("valid")]; - tensor var_3715_strides_0 = const()[name = string("op_3715_strides_0"), val = tensor([1, 1])]; - tensor var_3715_pad_0 = const()[name = string("op_3715_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3715_dilations_0 = const()[name = string("op_3715_dilations_0"), val = tensor([1, 1])]; - int32 var_3715_groups_0 = const()[name = string("op_3715_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126008832))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126000320))))[name = string("layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3715_cast_fp16 = conv(dilations = var_3715_dilations_0, groups = var_3715_groups_0, pad = var_3715_pad_0, pad_type = var_3715_pad_type_0, strides = var_3715_strides_0, weight = layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = string("op_3715_cast_fp16")]; - tensor value_19_cast_fp16 = add(x = var_3709_cast_fp16, y = var_3715_cast_fp16)[name = string("value_19_cast_fp16")]; - tensor var_3718_to_fp16 = const()[name = string("op_3718_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126139968)))]; - tensor query_39_cast_fp16 = add(x = query_37_cast_fp16, y = var_3718_to_fp16)[name = string("query_39_cast_fp16")]; - tensor var_3721_to_fp16 = const()[name = string("op_3721_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126142080)))]; - tensor q_with_bias_v_19_cast_fp16 = add(x = query_37_cast_fp16, y = var_3721_to_fp16)[name = string("q_with_bias_v_19_cast_fp16")]; - string var_3731_pad_type_0 = const()[name = string("op_3731_pad_type_0"), val = string("valid")]; - tensor var_3731_strides_0 = const()[name = string("op_3731_strides_0"), val = tensor([1, 1])]; - tensor var_3731_pad_0 = const()[name = string("op_3731_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3731_dilations_0 = const()[name = string("op_3731_dilations_0"), val = tensor([1, 1])]; - int32 var_3731_groups_0 = const()[name = string("op_3731_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126144192))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126537472))))[name = string("layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_3731_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3731_dilations_0, groups = var_3731_groups_0, pad = var_3731_pad_0, pad_type = var_3731_pad_type_0, strides = var_3731_strides_0, weight = layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_3731_cast_fp16")]; - string var_3737_pad_type_0 = const()[name = string("op_3737_pad_type_0"), val = string("valid")]; - tensor var_3737_strides_0 = const()[name = string("op_3737_strides_0"), val = tensor([1, 1])]; - tensor var_3737_pad_0 = const()[name = string("op_3737_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3737_dilations_0 = const()[name = string("op_3737_dilations_0"), val = tensor([1, 1])]; - int32 var_3737_groups_0 = const()[name = string("op_3737_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126572480))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126538560))))[name = string("layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3737_cast_fp16 = conv(dilations = var_3737_dilations_0, groups = var_3737_groups_0, pad = var_3737_pad_0, pad_type = var_3737_pad_type_0, strides = var_3737_strides_0, weight = layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_3737_cast_fp16")]; - tensor p_19_cast_fp16 = add(x = var_3731_cast_fp16, y = var_3737_cast_fp16)[name = string("p_19_cast_fp16")]; - tensor var_3741 = const()[name = string("op_3741"), val = tensor([1, 8, 128, 188])]; - tensor var_3742_cast_fp16 = reshape(shape = var_3741, x = q_with_bias_v_19_cast_fp16)[name = string("op_3742_cast_fp16")]; - tensor var_3743 = const()[name = string("op_3743"), val = tensor([1, 8, 128, -1])]; - tensor var_3744_cast_fp16 = reshape(shape = var_3743, x = p_19_cast_fp16)[name = string("op_3744_cast_fp16")]; - bool matrix_bd_73_transpose_x_0 = const()[name = string("matrix_bd_73_transpose_x_0"), val = bool(true)]; - bool matrix_bd_73_transpose_y_0 = const()[name = string("matrix_bd_73_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_73_cast_fp16 = matmul(transpose_x = matrix_bd_73_transpose_x_0, transpose_y = matrix_bd_73_transpose_y_0, x = var_3742_cast_fp16, y = var_3744_cast_fp16)[name = string("matrix_bd_73_cast_fp16")]; - tensor matrix_bd_75_pad_0 = const()[name = string("matrix_bd_75_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_75_mode_0 = const()[name = string("matrix_bd_75_mode_0"), val = string("constant")]; - fp16 const_109_to_fp16 = const()[name = string("const_109_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_75_cast_fp16 = pad(constant_val = const_109_to_fp16, mode = matrix_bd_75_mode_0, pad = matrix_bd_75_pad_0, x = matrix_bd_73_cast_fp16)[name = string("matrix_bd_75_cast_fp16")]; - tensor var_3753 = const()[name = string("op_3753"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_77_cast_fp16 = reshape(shape = var_3753, x = matrix_bd_75_cast_fp16)[name = string("matrix_bd_77_cast_fp16")]; - tensor var_3757_begin_0 = const()[name = string("op_3757_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_3757_end_0 = const()[name = string("op_3757_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_3757_end_mask_0 = const()[name = string("op_3757_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_3757_cast_fp16 = slice_by_index(begin = var_3757_begin_0, end = var_3757_end_0, end_mask = var_3757_end_mask_0, x = matrix_bd_77_cast_fp16)[name = string("op_3757_cast_fp16")]; - tensor var_3758 = const()[name = string("op_3758"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_79_cast_fp16 = reshape(shape = var_3758, x = var_3757_cast_fp16)[name = string("matrix_bd_79_cast_fp16")]; - tensor var_3763_begin_0 = const()[name = string("op_3763_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3763_end_0 = const()[name = string("op_3763_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_3763_end_mask_0 = const()[name = string("op_3763_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_3763_cast_fp16 = slice_by_index(begin = var_3763_begin_0, end = var_3763_end_0, end_mask = var_3763_end_mask_0, x = matrix_bd_79_cast_fp16)[name = string("op_3763_cast_fp16")]; - fp16 var_3764_to_fp16 = const()[name = string("op_3764_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_19_cast_fp16 = mul(x = var_3763_cast_fp16, y = var_3764_to_fp16)[name = string("qk_mask_19_cast_fp16")]; - tensor var_3768 = const()[name = string("op_3768"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_19_cast_fp16 = reshape(shape = var_3768, x = query_39_cast_fp16)[name = string("mh_q_19_cast_fp16")]; - fp16 var_3770_to_fp16 = const()[name = string("op_3770_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_3771_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_3770_to_fp16)[name = string("op_3771_cast_fp16")]; - tensor var_3774 = const()[name = string("op_3774"), val = tensor([1, 8, 128, 188])]; - tensor var_3775_cast_fp16 = reshape(shape = var_3774, x = key_19_cast_fp16)[name = string("op_3775_cast_fp16")]; - bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; - bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; - tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_3771_cast_fp16, y = var_3775_cast_fp16)[name = string("mh_w_37_cast_fp16")]; - tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = qk_mask_19_cast_fp16)[name = string("mh_w_39_cast_fp16")]; - tensor var_3779_cast_fp16 = softmax(axis = var_3566, x = mh_w_39_cast_fp16)[name = string("op_3779_cast_fp16")]; - tensor var_3780 = const()[name = string("op_3780"), val = tensor([1, 8, 128, 188])]; - tensor var_3781_cast_fp16 = reshape(shape = var_3780, x = value_19_cast_fp16)[name = string("op_3781_cast_fp16")]; - bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; - bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; - tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_3781_cast_fp16, y = var_3779_cast_fp16)[name = string("attn_19_cast_fp16")]; - tensor var_3784 = const()[name = string("op_3784"), val = tensor([1, 1024, 1, 188])]; - tensor input_257_cast_fp16 = reshape(shape = var_3784, x = attn_19_cast_fp16)[name = string("input_257_cast_fp16")]; - string var_3794_pad_type_0 = const()[name = string("op_3794_pad_type_0"), val = string("valid")]; - tensor var_3794_strides_0 = const()[name = string("op_3794_strides_0"), val = tensor([1, 1])]; - tensor var_3794_pad_0 = const()[name = string("op_3794_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3794_dilations_0 = const()[name = string("op_3794_dilations_0"), val = tensor([1, 1])]; - int32 var_3794_groups_0 = const()[name = string("op_3794_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126703616))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127096896))))[name = string("layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_3794_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3794_dilations_0, groups = var_3794_groups_0, pad = var_3794_pad_0, pad_type = var_3794_pad_type_0, strides = var_3794_strides_0, weight = layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_257_cast_fp16)[name = string("op_3794_cast_fp16")]; - string var_3800_pad_type_0 = const()[name = string("op_3800_pad_type_0"), val = string("valid")]; - tensor var_3800_strides_0 = const()[name = string("op_3800_strides_0"), val = tensor([1, 1])]; - tensor var_3800_pad_0 = const()[name = string("op_3800_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3800_dilations_0 = const()[name = string("op_3800_dilations_0"), val = tensor([1, 1])]; - int32 var_3800_groups_0 = const()[name = string("op_3800_groups_0"), val = int32(1)]; - tensor layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127106048))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127097984))))[name = string("layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3800_cast_fp16 = conv(dilations = var_3800_dilations_0, groups = var_3800_groups_0, pad = var_3800_pad_0, pad_type = var_3800_pad_type_0, strides = var_3800_strides_0, weight = layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_257_cast_fp16)[name = string("op_3800_cast_fp16")]; - tensor obj_41_cast_fp16 = add(x = var_3794_cast_fp16, y = var_3800_cast_fp16)[name = string("obj_41_cast_fp16")]; - tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_41_cast_fp16)[name = string("inputs_95_cast_fp16")]; - tensor out_95_axes_0 = const()[name = string("out_95_axes_0"), val = tensor([1])]; - fp16 var_3811_to_fp16 = const()[name = string("op_3811_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_3811_to_fp16, x = inputs_95_cast_fp16)[name = string("out_95_cast_fp16")]; - tensor input_259_gamma_0_to_fp16 = const()[name = string("input_259_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127237184)))]; - tensor input_259_beta_0_to_fp16 = const()[name = string("input_259_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127239296)))]; - fp16 input_259_epsilon_0_to_fp16 = const()[name = string("input_259_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_259_cast_fp16 = batch_norm(beta = input_259_beta_0_to_fp16, epsilon = input_259_epsilon_0_to_fp16, gamma = input_259_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_95_cast_fp16)[name = string("input_259_cast_fp16")]; - string var_3832_pad_type_0 = const()[name = string("op_3832_pad_type_0"), val = string("valid")]; - tensor var_3832_strides_0 = const()[name = string("op_3832_strides_0"), val = tensor([1, 1])]; - tensor var_3832_pad_0 = const()[name = string("op_3832_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3832_dilations_0 = const()[name = string("op_3832_dilations_0"), val = tensor([1, 1])]; - int32 var_3832_groups_0 = const()[name = string("op_3832_groups_0"), val = int32(1)]; - tensor layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127241408))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128027904))))[name = string("layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3832_cast_fp16 = conv(dilations = var_3832_dilations_0, groups = var_3832_groups_0, pad = var_3832_pad_0, pad_type = var_3832_pad_type_0, strides = var_3832_strides_0, weight = layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_259_cast_fp16)[name = string("op_3832_cast_fp16")]; - string var_3838_pad_type_0 = const()[name = string("op_3838_pad_type_0"), val = string("valid")]; - tensor var_3838_strides_0 = const()[name = string("op_3838_strides_0"), val = tensor([1, 1])]; - tensor var_3838_pad_0 = const()[name = string("op_3838_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3838_dilations_0 = const()[name = string("op_3838_dilations_0"), val = tensor([1, 1])]; - int32 var_3838_groups_0 = const()[name = string("op_3838_groups_0"), val = int32(1)]; - tensor layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128046784))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128030016))))[name = string("layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3838_cast_fp16 = conv(dilations = var_3838_dilations_0, groups = var_3838_groups_0, pad = var_3838_pad_0, pad_type = var_3838_pad_type_0, strides = var_3838_strides_0, weight = layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_259_cast_fp16)[name = string("op_3838_cast_fp16")]; - tensor input_261_cast_fp16 = add(x = var_3832_cast_fp16, y = var_3838_cast_fp16)[name = string("input_261_cast_fp16")]; - int32 input_263_split_num_splits_0 = const()[name = string("input_263_split_num_splits_0"), val = int32(2)]; - int32 input_263_split_axis_0 = const()[name = string("input_263_split_axis_0"), val = int32(1)]; - tensor input_263_split_cast_fp16_0, tensor input_263_split_cast_fp16_1 = split(axis = input_263_split_axis_0, num_splits = input_263_split_num_splits_0, x = input_261_cast_fp16)[name = string("input_263_split_cast_fp16")]; - tensor input_263_split_1_sigmoid_cast_fp16 = sigmoid(x = input_263_split_cast_fp16_1)[name = string("input_263_split_1_sigmoid_cast_fp16")]; - tensor input_263_cast_fp16 = mul(x = input_263_split_cast_fp16_0, y = input_263_split_1_sigmoid_cast_fp16)[name = string("input_263_cast_fp16")]; - string input_265_pad_type_0 = const()[name = string("input_265_pad_type_0"), val = string("custom")]; - tensor input_265_pad_0 = const()[name = string("input_265_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_265_groups_0 = const()[name = string("input_265_groups_0"), val = int32(1024)]; - tensor input_265_strides_0 = const()[name = string("input_265_strides_0"), val = tensor([1, 1])]; - tensor input_265_dilations_0 = const()[name = string("input_265_dilations_0"), val = tensor([1, 1])]; - tensor const_286_to_fp16 = const()[name = string("const_286_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128308992)))]; - tensor const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128327488)))]; - tensor input_267_cast_fp16 = conv(bias = const_287_to_fp16, dilations = input_265_dilations_0, groups = input_265_groups_0, pad = input_265_pad_0, pad_type = input_265_pad_type_0, strides = input_265_strides_0, weight = const_286_to_fp16, x = input_263_cast_fp16)[name = string("input_267_cast_fp16")]; - tensor input_269_cast_fp16 = silu(x = input_267_cast_fp16)[name = string("input_269_cast_fp16")]; - string var_3860_pad_type_0 = const()[name = string("op_3860_pad_type_0"), val = string("valid")]; - tensor var_3860_strides_0 = const()[name = string("op_3860_strides_0"), val = tensor([1, 1])]; - tensor var_3860_pad_0 = const()[name = string("op_3860_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3860_dilations_0 = const()[name = string("op_3860_dilations_0"), val = tensor([1, 1])]; - int32 var_3860_groups_0 = const()[name = string("op_3860_groups_0"), val = int32(1)]; - tensor layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128329600))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128722880))))[name = string("layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3860_cast_fp16 = conv(dilations = var_3860_dilations_0, groups = var_3860_groups_0, pad = var_3860_pad_0, pad_type = var_3860_pad_type_0, strides = var_3860_strides_0, weight = layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_269_cast_fp16)[name = string("op_3860_cast_fp16")]; - string var_3866_pad_type_0 = const()[name = string("op_3866_pad_type_0"), val = string("valid")]; - tensor var_3866_strides_0 = const()[name = string("op_3866_strides_0"), val = tensor([1, 1])]; - tensor var_3866_pad_0 = const()[name = string("op_3866_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3866_dilations_0 = const()[name = string("op_3866_dilations_0"), val = tensor([1, 1])]; - int32 var_3866_groups_0 = const()[name = string("op_3866_groups_0"), val = int32(1)]; - tensor layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128733056))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128723968))))[name = string("layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3866_cast_fp16 = conv(dilations = var_3866_dilations_0, groups = var_3866_groups_0, pad = var_3866_pad_0, pad_type = var_3866_pad_type_0, strides = var_3866_strides_0, weight = layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_269_cast_fp16)[name = string("op_3866_cast_fp16")]; - tensor x_59_cast_fp16 = add(x = var_3860_cast_fp16, y = var_3866_cast_fp16)[name = string("x_59_cast_fp16")]; - tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = x_59_cast_fp16)[name = string("inputs_97_cast_fp16")]; - tensor out_97_axes_0 = const()[name = string("out_97_axes_0"), val = tensor([1])]; - fp16 var_3877_to_fp16 = const()[name = string("op_3877_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3877_to_fp16, x = inputs_97_cast_fp16)[name = string("out_97_cast_fp16")]; - tensor input_271_gamma_0_to_fp16 = const()[name = string("input_271_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128864192)))]; - tensor input_271_beta_0_to_fp16 = const()[name = string("input_271_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128866304)))]; - fp16 input_271_epsilon_0_to_fp16 = const()[name = string("input_271_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_271_cast_fp16 = batch_norm(beta = input_271_beta_0_to_fp16, epsilon = input_271_epsilon_0_to_fp16, gamma = input_271_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_97_cast_fp16)[name = string("input_271_cast_fp16")]; - string var_3897_pad_type_0 = const()[name = string("op_3897_pad_type_0"), val = string("valid")]; - tensor var_3897_strides_0 = const()[name = string("op_3897_strides_0"), val = tensor([1, 1])]; - tensor var_3897_pad_0 = const()[name = string("op_3897_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3897_dilations_0 = const()[name = string("op_3897_dilations_0"), val = tensor([1, 1])]; - int32 var_3897_groups_0 = const()[name = string("op_3897_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128868416))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130441344))))[name = string("layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3897_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3897_dilations_0, groups = var_3897_groups_0, pad = var_3897_pad_0, pad_type = var_3897_pad_type_0, strides = var_3897_strides_0, weight = layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = string("op_3897_cast_fp16")]; - string var_3903_pad_type_0 = const()[name = string("op_3903_pad_type_0"), val = string("valid")]; - tensor var_3903_strides_0 = const()[name = string("op_3903_strides_0"), val = tensor([1, 1])]; - tensor var_3903_pad_0 = const()[name = string("op_3903_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3903_dilations_0 = const()[name = string("op_3903_dilations_0"), val = tensor([1, 1])]; - int32 var_3903_groups_0 = const()[name = string("op_3903_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130490688))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130445504))))[name = string("layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3903_cast_fp16 = conv(dilations = var_3903_dilations_0, groups = var_3903_groups_0, pad = var_3903_pad_0, pad_type = var_3903_pad_type_0, strides = var_3903_strides_0, weight = layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_271_cast_fp16)[name = string("op_3903_cast_fp16")]; - tensor input_273_cast_fp16 = add(x = var_3897_cast_fp16, y = var_3903_cast_fp16)[name = string("input_273_cast_fp16")]; - tensor input_275_cast_fp16 = silu(x = input_273_cast_fp16)[name = string("input_275_cast_fp16")]; - string var_3914_pad_type_0 = const()[name = string("op_3914_pad_type_0"), val = string("valid")]; - tensor var_3914_strides_0 = const()[name = string("op_3914_strides_0"), val = tensor([1, 1])]; - tensor var_3914_pad_0 = const()[name = string("op_3914_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3914_dilations_0 = const()[name = string("op_3914_dilations_0"), val = tensor([1, 1])]; - int32 var_3914_groups_0 = const()[name = string("op_3914_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131015040))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132587968))))[name = string("layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_3914_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3914_dilations_0, groups = var_3914_groups_0, pad = var_3914_pad_0, pad_type = var_3914_pad_type_0, strides = var_3914_strides_0, weight = layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_275_cast_fp16)[name = string("op_3914_cast_fp16")]; - string var_3920_pad_type_0 = const()[name = string("op_3920_pad_type_0"), val = string("valid")]; - tensor var_3920_strides_0 = const()[name = string("op_3920_strides_0"), val = tensor([1, 1])]; - tensor var_3920_pad_0 = const()[name = string("op_3920_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3920_dilations_0 = const()[name = string("op_3920_dilations_0"), val = tensor([1, 1])]; - int32 var_3920_groups_0 = const()[name = string("op_3920_groups_0"), val = int32(1)]; - tensor layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132631296))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132589056))))[name = string("layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_3920_cast_fp16 = conv(dilations = var_3920_dilations_0, groups = var_3920_groups_0, pad = var_3920_pad_0, pad_type = var_3920_pad_type_0, strides = var_3920_strides_0, weight = layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_275_cast_fp16)[name = string("op_3920_cast_fp16")]; - tensor x_61_cast_fp16 = add(x = var_3914_cast_fp16, y = var_3920_cast_fp16)[name = string("x_61_cast_fp16")]; - fp16 var_3922_to_fp16 = const()[name = string("op_3922_to_fp16"), val = fp16(0x1p-1)]; - tensor var_3923_cast_fp16 = mul(x = x_61_cast_fp16, y = var_3922_to_fp16)[name = string("op_3923_cast_fp16")]; - tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = var_3923_cast_fp16)[name = string("inputs_99_cast_fp16")]; - tensor out_99_axes_0 = const()[name = string("out_99_axes_0"), val = tensor([1])]; - fp16 var_3933_to_fp16 = const()[name = string("op_3933_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3933_to_fp16, x = inputs_99_cast_fp16)[name = string("out_99_cast_fp16")]; - tensor inputs_101_gamma_0_to_fp16 = const()[name = string("inputs_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133155648)))]; - tensor inputs_101_beta_0_to_fp16 = const()[name = string("inputs_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133157760)))]; - fp16 inputs_101_epsilon_0_to_fp16 = const()[name = string("inputs_101_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_101_cast_fp16 = batch_norm(beta = inputs_101_beta_0_to_fp16, epsilon = inputs_101_epsilon_0_to_fp16, gamma = inputs_101_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_99_cast_fp16)[name = string("inputs_101_cast_fp16")]; - int32 var_3947 = const()[name = string("op_3947"), val = int32(3)]; - tensor out_101_axes_0 = const()[name = string("out_101_axes_0"), val = tensor([1])]; - fp16 var_3978_to_fp16 = const()[name = string("op_3978_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3978_to_fp16, x = inputs_101_cast_fp16)[name = string("out_101_cast_fp16")]; - tensor input_277_gamma_0_to_fp16 = const()[name = string("input_277_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133159872)))]; - tensor input_277_beta_0_to_fp16 = const()[name = string("input_277_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133161984)))]; - fp16 input_277_epsilon_0_to_fp16 = const()[name = string("input_277_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_277_cast_fp16 = batch_norm(beta = input_277_beta_0_to_fp16, epsilon = input_277_epsilon_0_to_fp16, gamma = input_277_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_101_cast_fp16)[name = string("input_277_cast_fp16")]; - string var_3998_pad_type_0 = const()[name = string("op_3998_pad_type_0"), val = string("valid")]; - tensor var_3998_strides_0 = const()[name = string("op_3998_strides_0"), val = tensor([1, 1])]; - tensor var_3998_pad_0 = const()[name = string("op_3998_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_3998_dilations_0 = const()[name = string("op_3998_dilations_0"), val = tensor([1, 1])]; - int32 var_3998_groups_0 = const()[name = string("op_3998_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133164096))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134737024))))[name = string("layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_3998_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3998_dilations_0, groups = var_3998_groups_0, pad = var_3998_pad_0, pad_type = var_3998_pad_type_0, strides = var_3998_strides_0, weight = layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_277_cast_fp16)[name = string("op_3998_cast_fp16")]; - string var_4004_pad_type_0 = const()[name = string("op_4004_pad_type_0"), val = string("valid")]; - tensor var_4004_strides_0 = const()[name = string("op_4004_strides_0"), val = tensor([1, 1])]; - tensor var_4004_pad_0 = const()[name = string("op_4004_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4004_dilations_0 = const()[name = string("op_4004_dilations_0"), val = tensor([1, 1])]; - int32 var_4004_groups_0 = const()[name = string("op_4004_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134791936))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134741184))))[name = string("layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4004_cast_fp16 = conv(dilations = var_4004_dilations_0, groups = var_4004_groups_0, pad = var_4004_pad_0, pad_type = var_4004_pad_type_0, strides = var_4004_strides_0, weight = layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_277_cast_fp16)[name = string("op_4004_cast_fp16")]; - tensor input_279_cast_fp16 = add(x = var_3998_cast_fp16, y = var_4004_cast_fp16)[name = string("input_279_cast_fp16")]; - tensor input_281_cast_fp16 = silu(x = input_279_cast_fp16)[name = string("input_281_cast_fp16")]; - string var_4015_pad_type_0 = const()[name = string("op_4015_pad_type_0"), val = string("valid")]; - tensor var_4015_strides_0 = const()[name = string("op_4015_strides_0"), val = tensor([1, 1])]; - tensor var_4015_pad_0 = const()[name = string("op_4015_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4015_dilations_0 = const()[name = string("op_4015_dilations_0"), val = tensor([1, 1])]; - int32 var_4015_groups_0 = const()[name = string("op_4015_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135316288))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136889216))))[name = string("layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_4015_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4015_dilations_0, groups = var_4015_groups_0, pad = var_4015_pad_0, pad_type = var_4015_pad_type_0, strides = var_4015_strides_0, weight = layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_281_cast_fp16)[name = string("op_4015_cast_fp16")]; - string var_4021_pad_type_0 = const()[name = string("op_4021_pad_type_0"), val = string("valid")]; - tensor var_4021_strides_0 = const()[name = string("op_4021_strides_0"), val = tensor([1, 1])]; - tensor var_4021_pad_0 = const()[name = string("op_4021_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4021_dilations_0 = const()[name = string("op_4021_dilations_0"), val = tensor([1, 1])]; - int32 var_4021_groups_0 = const()[name = string("op_4021_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136938368))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136890304))))[name = string("layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4021_cast_fp16 = conv(dilations = var_4021_dilations_0, groups = var_4021_groups_0, pad = var_4021_pad_0, pad_type = var_4021_pad_type_0, strides = var_4021_strides_0, weight = layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_281_cast_fp16)[name = string("op_4021_cast_fp16")]; - tensor x_63_cast_fp16 = add(x = var_4015_cast_fp16, y = var_4021_cast_fp16)[name = string("x_63_cast_fp16")]; - fp16 var_4023_to_fp16 = const()[name = string("op_4023_to_fp16"), val = fp16(0x1p-1)]; - tensor var_4024_cast_fp16 = mul(x = x_63_cast_fp16, y = var_4023_to_fp16)[name = string("op_4024_cast_fp16")]; - tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = var_4024_cast_fp16)[name = string("inputs_103_cast_fp16")]; - tensor out_103_axes_0 = const()[name = string("out_103_axes_0"), val = tensor([1])]; - fp16 var_4034_to_fp16 = const()[name = string("op_4034_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_4034_to_fp16, x = inputs_103_cast_fp16)[name = string("out_103_cast_fp16")]; - tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137462720)))]; - tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137464832)))]; - fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_103_cast_fp16)[name = string("obj_43_cast_fp16")]; - string var_4059_pad_type_0 = const()[name = string("op_4059_pad_type_0"), val = string("valid")]; - tensor var_4059_strides_0 = const()[name = string("op_4059_strides_0"), val = tensor([1, 1])]; - tensor var_4059_pad_0 = const()[name = string("op_4059_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4059_dilations_0 = const()[name = string("op_4059_dilations_0"), val = tensor([1, 1])]; - int32 var_4059_groups_0 = const()[name = string("op_4059_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137466944))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137860224))))[name = string("layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4059_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4059_dilations_0, groups = var_4059_groups_0, pad = var_4059_pad_0, pad_type = var_4059_pad_type_0, strides = var_4059_strides_0, weight = layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = string("op_4059_cast_fp16")]; - string var_4065_pad_type_0 = const()[name = string("op_4065_pad_type_0"), val = string("valid")]; - tensor var_4065_strides_0 = const()[name = string("op_4065_strides_0"), val = tensor([1, 1])]; - tensor var_4065_pad_0 = const()[name = string("op_4065_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4065_dilations_0 = const()[name = string("op_4065_dilations_0"), val = tensor([1, 1])]; - int32 var_4065_groups_0 = const()[name = string("op_4065_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137877184))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137861312))))[name = string("layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4065_cast_fp16 = conv(dilations = var_4065_dilations_0, groups = var_4065_groups_0, pad = var_4065_pad_0, pad_type = var_4065_pad_type_0, strides = var_4065_strides_0, weight = layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = string("op_4065_cast_fp16")]; - tensor query_41_cast_fp16 = add(x = var_4059_cast_fp16, y = var_4065_cast_fp16)[name = string("query_41_cast_fp16")]; - string var_4074_pad_type_0 = const()[name = string("op_4074_pad_type_0"), val = string("valid")]; - tensor var_4074_strides_0 = const()[name = string("op_4074_strides_0"), val = tensor([1, 1])]; - tensor var_4074_pad_0 = const()[name = string("op_4074_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4074_dilations_0 = const()[name = string("op_4074_dilations_0"), val = tensor([1, 1])]; - int32 var_4074_groups_0 = const()[name = string("op_4074_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138008320))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138401600))))[name = string("layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4074_cast_fp16 = conv(dilations = var_4074_dilations_0, groups = var_4074_groups_0, pad = var_4074_pad_0, pad_type = var_4074_pad_type_0, strides = var_4074_strides_0, weight = layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = string("op_4074_cast_fp16")]; - string var_4080_pad_type_0 = const()[name = string("op_4080_pad_type_0"), val = string("valid")]; - tensor var_4080_strides_0 = const()[name = string("op_4080_strides_0"), val = tensor([1, 1])]; - tensor var_4080_pad_0 = const()[name = string("op_4080_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4080_dilations_0 = const()[name = string("op_4080_dilations_0"), val = tensor([1, 1])]; - int32 var_4080_groups_0 = const()[name = string("op_4080_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138421440))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138402688))))[name = string("layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4080_cast_fp16 = conv(dilations = var_4080_dilations_0, groups = var_4080_groups_0, pad = var_4080_pad_0, pad_type = var_4080_pad_type_0, strides = var_4080_strides_0, weight = layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = string("op_4080_cast_fp16")]; - tensor key_21_cast_fp16 = add(x = var_4074_cast_fp16, y = var_4080_cast_fp16)[name = string("key_21_cast_fp16")]; - string var_4090_pad_type_0 = const()[name = string("op_4090_pad_type_0"), val = string("valid")]; - tensor var_4090_strides_0 = const()[name = string("op_4090_strides_0"), val = tensor([1, 1])]; - tensor var_4090_pad_0 = const()[name = string("op_4090_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4090_dilations_0 = const()[name = string("op_4090_dilations_0"), val = tensor([1, 1])]; - int32 var_4090_groups_0 = const()[name = string("op_4090_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138552576))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138945856))))[name = string("layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4090_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4090_dilations_0, groups = var_4090_groups_0, pad = var_4090_pad_0, pad_type = var_4090_pad_type_0, strides = var_4090_strides_0, weight = layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = string("op_4090_cast_fp16")]; - string var_4096_pad_type_0 = const()[name = string("op_4096_pad_type_0"), val = string("valid")]; - tensor var_4096_strides_0 = const()[name = string("op_4096_strides_0"), val = tensor([1, 1])]; - tensor var_4096_pad_0 = const()[name = string("op_4096_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4096_dilations_0 = const()[name = string("op_4096_dilations_0"), val = tensor([1, 1])]; - int32 var_4096_groups_0 = const()[name = string("op_4096_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138955392))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138946944))))[name = string("layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4096_cast_fp16 = conv(dilations = var_4096_dilations_0, groups = var_4096_groups_0, pad = var_4096_pad_0, pad_type = var_4096_pad_type_0, strides = var_4096_strides_0, weight = layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = string("op_4096_cast_fp16")]; - tensor value_21_cast_fp16 = add(x = var_4090_cast_fp16, y = var_4096_cast_fp16)[name = string("value_21_cast_fp16")]; - tensor var_4099_to_fp16 = const()[name = string("op_4099_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139086528)))]; - tensor query_43_cast_fp16 = add(x = query_41_cast_fp16, y = var_4099_to_fp16)[name = string("query_43_cast_fp16")]; - tensor var_4102_to_fp16 = const()[name = string("op_4102_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139088640)))]; - tensor q_with_bias_v_21_cast_fp16 = add(x = query_41_cast_fp16, y = var_4102_to_fp16)[name = string("q_with_bias_v_21_cast_fp16")]; - string var_4112_pad_type_0 = const()[name = string("op_4112_pad_type_0"), val = string("valid")]; - tensor var_4112_strides_0 = const()[name = string("op_4112_strides_0"), val = tensor([1, 1])]; - tensor var_4112_pad_0 = const()[name = string("op_4112_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4112_dilations_0 = const()[name = string("op_4112_dilations_0"), val = tensor([1, 1])]; - int32 var_4112_groups_0 = const()[name = string("op_4112_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139090752))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139484032))))[name = string("layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_4112_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4112_dilations_0, groups = var_4112_groups_0, pad = var_4112_pad_0, pad_type = var_4112_pad_type_0, strides = var_4112_strides_0, weight = layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_4112_cast_fp16")]; - string var_4118_pad_type_0 = const()[name = string("op_4118_pad_type_0"), val = string("valid")]; - tensor var_4118_strides_0 = const()[name = string("op_4118_strides_0"), val = tensor([1, 1])]; - tensor var_4118_pad_0 = const()[name = string("op_4118_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4118_dilations_0 = const()[name = string("op_4118_dilations_0"), val = tensor([1, 1])]; - int32 var_4118_groups_0 = const()[name = string("op_4118_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139516608))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139485120))))[name = string("layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4118_cast_fp16 = conv(dilations = var_4118_dilations_0, groups = var_4118_groups_0, pad = var_4118_pad_0, pad_type = var_4118_pad_type_0, strides = var_4118_strides_0, weight = layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_4118_cast_fp16")]; - tensor p_21_cast_fp16 = add(x = var_4112_cast_fp16, y = var_4118_cast_fp16)[name = string("p_21_cast_fp16")]; - tensor var_4122 = const()[name = string("op_4122"), val = tensor([1, 8, 128, 188])]; - tensor var_4123_cast_fp16 = reshape(shape = var_4122, x = q_with_bias_v_21_cast_fp16)[name = string("op_4123_cast_fp16")]; - tensor var_4124 = const()[name = string("op_4124"), val = tensor([1, 8, 128, -1])]; - tensor var_4125_cast_fp16 = reshape(shape = var_4124, x = p_21_cast_fp16)[name = string("op_4125_cast_fp16")]; - bool matrix_bd_81_transpose_x_0 = const()[name = string("matrix_bd_81_transpose_x_0"), val = bool(true)]; - bool matrix_bd_81_transpose_y_0 = const()[name = string("matrix_bd_81_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_81_cast_fp16 = matmul(transpose_x = matrix_bd_81_transpose_x_0, transpose_y = matrix_bd_81_transpose_y_0, x = var_4123_cast_fp16, y = var_4125_cast_fp16)[name = string("matrix_bd_81_cast_fp16")]; - tensor matrix_bd_83_pad_0 = const()[name = string("matrix_bd_83_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_83_mode_0 = const()[name = string("matrix_bd_83_mode_0"), val = string("constant")]; - fp16 const_120_to_fp16 = const()[name = string("const_120_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_83_cast_fp16 = pad(constant_val = const_120_to_fp16, mode = matrix_bd_83_mode_0, pad = matrix_bd_83_pad_0, x = matrix_bd_81_cast_fp16)[name = string("matrix_bd_83_cast_fp16")]; - tensor var_4134 = const()[name = string("op_4134"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_85_cast_fp16 = reshape(shape = var_4134, x = matrix_bd_83_cast_fp16)[name = string("matrix_bd_85_cast_fp16")]; - tensor var_4138_begin_0 = const()[name = string("op_4138_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_4138_end_0 = const()[name = string("op_4138_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_4138_end_mask_0 = const()[name = string("op_4138_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_4138_cast_fp16 = slice_by_index(begin = var_4138_begin_0, end = var_4138_end_0, end_mask = var_4138_end_mask_0, x = matrix_bd_85_cast_fp16)[name = string("op_4138_cast_fp16")]; - tensor var_4139 = const()[name = string("op_4139"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_87_cast_fp16 = reshape(shape = var_4139, x = var_4138_cast_fp16)[name = string("matrix_bd_87_cast_fp16")]; - tensor var_4144_begin_0 = const()[name = string("op_4144_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4144_end_0 = const()[name = string("op_4144_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_4144_end_mask_0 = const()[name = string("op_4144_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_4144_cast_fp16 = slice_by_index(begin = var_4144_begin_0, end = var_4144_end_0, end_mask = var_4144_end_mask_0, x = matrix_bd_87_cast_fp16)[name = string("op_4144_cast_fp16")]; - fp16 var_4145_to_fp16 = const()[name = string("op_4145_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_21_cast_fp16 = mul(x = var_4144_cast_fp16, y = var_4145_to_fp16)[name = string("qk_mask_21_cast_fp16")]; - tensor var_4149 = const()[name = string("op_4149"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_21_cast_fp16 = reshape(shape = var_4149, x = query_43_cast_fp16)[name = string("mh_q_21_cast_fp16")]; - fp16 var_4151_to_fp16 = const()[name = string("op_4151_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_4152_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_4151_to_fp16)[name = string("op_4152_cast_fp16")]; - tensor var_4155 = const()[name = string("op_4155"), val = tensor([1, 8, 128, 188])]; - tensor var_4156_cast_fp16 = reshape(shape = var_4155, x = key_21_cast_fp16)[name = string("op_4156_cast_fp16")]; - bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; - bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; - tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_4152_cast_fp16, y = var_4156_cast_fp16)[name = string("mh_w_41_cast_fp16")]; - tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = qk_mask_21_cast_fp16)[name = string("mh_w_43_cast_fp16")]; - tensor var_4160_cast_fp16 = softmax(axis = var_3947, x = mh_w_43_cast_fp16)[name = string("op_4160_cast_fp16")]; - tensor var_4161 = const()[name = string("op_4161"), val = tensor([1, 8, 128, 188])]; - tensor var_4162_cast_fp16 = reshape(shape = var_4161, x = value_21_cast_fp16)[name = string("op_4162_cast_fp16")]; - bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; - bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; - tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_4162_cast_fp16, y = var_4160_cast_fp16)[name = string("attn_21_cast_fp16")]; - tensor var_4165 = const()[name = string("op_4165"), val = tensor([1, 1024, 1, 188])]; - tensor input_283_cast_fp16 = reshape(shape = var_4165, x = attn_21_cast_fp16)[name = string("input_283_cast_fp16")]; - string var_4175_pad_type_0 = const()[name = string("op_4175_pad_type_0"), val = string("valid")]; - tensor var_4175_strides_0 = const()[name = string("op_4175_strides_0"), val = tensor([1, 1])]; - tensor var_4175_pad_0 = const()[name = string("op_4175_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4175_dilations_0 = const()[name = string("op_4175_dilations_0"), val = tensor([1, 1])]; - int32 var_4175_groups_0 = const()[name = string("op_4175_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139647744))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140041024))))[name = string("layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4175_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4175_dilations_0, groups = var_4175_groups_0, pad = var_4175_pad_0, pad_type = var_4175_pad_type_0, strides = var_4175_strides_0, weight = layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_283_cast_fp16)[name = string("op_4175_cast_fp16")]; - string var_4181_pad_type_0 = const()[name = string("op_4181_pad_type_0"), val = string("valid")]; - tensor var_4181_strides_0 = const()[name = string("op_4181_strides_0"), val = tensor([1, 1])]; - tensor var_4181_pad_0 = const()[name = string("op_4181_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4181_dilations_0 = const()[name = string("op_4181_dilations_0"), val = tensor([1, 1])]; - int32 var_4181_groups_0 = const()[name = string("op_4181_groups_0"), val = int32(1)]; - tensor layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140051584))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140042112))))[name = string("layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4181_cast_fp16 = conv(dilations = var_4181_dilations_0, groups = var_4181_groups_0, pad = var_4181_pad_0, pad_type = var_4181_pad_type_0, strides = var_4181_strides_0, weight = layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_283_cast_fp16)[name = string("op_4181_cast_fp16")]; - tensor obj_45_cast_fp16 = add(x = var_4175_cast_fp16, y = var_4181_cast_fp16)[name = string("obj_45_cast_fp16")]; - tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = obj_45_cast_fp16)[name = string("inputs_105_cast_fp16")]; - tensor out_105_axes_0 = const()[name = string("out_105_axes_0"), val = tensor([1])]; - fp16 var_4192_to_fp16 = const()[name = string("op_4192_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_4192_to_fp16, x = inputs_105_cast_fp16)[name = string("out_105_cast_fp16")]; - tensor input_285_gamma_0_to_fp16 = const()[name = string("input_285_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140182720)))]; - tensor input_285_beta_0_to_fp16 = const()[name = string("input_285_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140184832)))]; - fp16 input_285_epsilon_0_to_fp16 = const()[name = string("input_285_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_285_cast_fp16 = batch_norm(beta = input_285_beta_0_to_fp16, epsilon = input_285_epsilon_0_to_fp16, gamma = input_285_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_105_cast_fp16)[name = string("input_285_cast_fp16")]; - string var_4213_pad_type_0 = const()[name = string("op_4213_pad_type_0"), val = string("valid")]; - tensor var_4213_strides_0 = const()[name = string("op_4213_strides_0"), val = tensor([1, 1])]; - tensor var_4213_pad_0 = const()[name = string("op_4213_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4213_dilations_0 = const()[name = string("op_4213_dilations_0"), val = tensor([1, 1])]; - int32 var_4213_groups_0 = const()[name = string("op_4213_groups_0"), val = int32(1)]; - tensor layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140186944))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140973440))))[name = string("layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_4213_cast_fp16 = conv(dilations = var_4213_dilations_0, groups = var_4213_groups_0, pad = var_4213_pad_0, pad_type = var_4213_pad_type_0, strides = var_4213_strides_0, weight = layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = string("op_4213_cast_fp16")]; - string var_4219_pad_type_0 = const()[name = string("op_4219_pad_type_0"), val = string("valid")]; - tensor var_4219_strides_0 = const()[name = string("op_4219_strides_0"), val = tensor([1, 1])]; - tensor var_4219_pad_0 = const()[name = string("op_4219_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4219_dilations_0 = const()[name = string("op_4219_dilations_0"), val = tensor([1, 1])]; - int32 var_4219_groups_0 = const()[name = string("op_4219_groups_0"), val = int32(1)]; - tensor layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140992768))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140975552))))[name = string("layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4219_cast_fp16 = conv(dilations = var_4219_dilations_0, groups = var_4219_groups_0, pad = var_4219_pad_0, pad_type = var_4219_pad_type_0, strides = var_4219_strides_0, weight = layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_285_cast_fp16)[name = string("op_4219_cast_fp16")]; - tensor input_287_cast_fp16 = add(x = var_4213_cast_fp16, y = var_4219_cast_fp16)[name = string("input_287_cast_fp16")]; - int32 input_289_split_num_splits_0 = const()[name = string("input_289_split_num_splits_0"), val = int32(2)]; - int32 input_289_split_axis_0 = const()[name = string("input_289_split_axis_0"), val = int32(1)]; - tensor input_289_split_cast_fp16_0, tensor input_289_split_cast_fp16_1 = split(axis = input_289_split_axis_0, num_splits = input_289_split_num_splits_0, x = input_287_cast_fp16)[name = string("input_289_split_cast_fp16")]; - tensor input_289_split_1_sigmoid_cast_fp16 = sigmoid(x = input_289_split_cast_fp16_1)[name = string("input_289_split_1_sigmoid_cast_fp16")]; - tensor input_289_cast_fp16 = mul(x = input_289_split_cast_fp16_0, y = input_289_split_1_sigmoid_cast_fp16)[name = string("input_289_cast_fp16")]; - string input_291_pad_type_0 = const()[name = string("input_291_pad_type_0"), val = string("custom")]; - tensor input_291_pad_0 = const()[name = string("input_291_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_291_groups_0 = const()[name = string("input_291_groups_0"), val = int32(1024)]; - tensor input_291_strides_0 = const()[name = string("input_291_strides_0"), val = tensor([1, 1])]; - tensor input_291_dilations_0 = const()[name = string("input_291_dilations_0"), val = tensor([1, 1])]; - tensor const_288_to_fp16 = const()[name = string("const_288_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141254976)))]; - tensor const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141273472)))]; - tensor input_293_cast_fp16 = conv(bias = const_289_to_fp16, dilations = input_291_dilations_0, groups = input_291_groups_0, pad = input_291_pad_0, pad_type = input_291_pad_type_0, strides = input_291_strides_0, weight = const_288_to_fp16, x = input_289_cast_fp16)[name = string("input_293_cast_fp16")]; - tensor input_295_cast_fp16 = silu(x = input_293_cast_fp16)[name = string("input_295_cast_fp16")]; - string var_4241_pad_type_0 = const()[name = string("op_4241_pad_type_0"), val = string("valid")]; - tensor var_4241_strides_0 = const()[name = string("op_4241_strides_0"), val = tensor([1, 1])]; - tensor var_4241_pad_0 = const()[name = string("op_4241_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4241_dilations_0 = const()[name = string("op_4241_dilations_0"), val = tensor([1, 1])]; - int32 var_4241_groups_0 = const()[name = string("op_4241_groups_0"), val = int32(1)]; - tensor layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141275584))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141668864))))[name = string("layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_4241_cast_fp16 = conv(dilations = var_4241_dilations_0, groups = var_4241_groups_0, pad = var_4241_pad_0, pad_type = var_4241_pad_type_0, strides = var_4241_strides_0, weight = layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_295_cast_fp16)[name = string("op_4241_cast_fp16")]; - string var_4247_pad_type_0 = const()[name = string("op_4247_pad_type_0"), val = string("valid")]; - tensor var_4247_strides_0 = const()[name = string("op_4247_strides_0"), val = tensor([1, 1])]; - tensor var_4247_pad_0 = const()[name = string("op_4247_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4247_dilations_0 = const()[name = string("op_4247_dilations_0"), val = tensor([1, 1])]; - int32 var_4247_groups_0 = const()[name = string("op_4247_groups_0"), val = int32(1)]; - tensor layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141679488))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141669952))))[name = string("layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4247_cast_fp16 = conv(dilations = var_4247_dilations_0, groups = var_4247_groups_0, pad = var_4247_pad_0, pad_type = var_4247_pad_type_0, strides = var_4247_strides_0, weight = layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_295_cast_fp16)[name = string("op_4247_cast_fp16")]; - tensor x_65_cast_fp16 = add(x = var_4241_cast_fp16, y = var_4247_cast_fp16)[name = string("x_65_cast_fp16")]; - tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = x_65_cast_fp16)[name = string("inputs_107_cast_fp16")]; - tensor out_107_axes_0 = const()[name = string("out_107_axes_0"), val = tensor([1])]; - fp16 var_4258_to_fp16 = const()[name = string("op_4258_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_4258_to_fp16, x = inputs_107_cast_fp16)[name = string("out_107_cast_fp16")]; - tensor input_297_gamma_0_to_fp16 = const()[name = string("input_297_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141810624)))]; - tensor input_297_beta_0_to_fp16 = const()[name = string("input_297_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141812736)))]; - fp16 input_297_epsilon_0_to_fp16 = const()[name = string("input_297_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_297_cast_fp16 = batch_norm(beta = input_297_beta_0_to_fp16, epsilon = input_297_epsilon_0_to_fp16, gamma = input_297_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_107_cast_fp16)[name = string("input_297_cast_fp16")]; - string var_4278_pad_type_0 = const()[name = string("op_4278_pad_type_0"), val = string("valid")]; - tensor var_4278_strides_0 = const()[name = string("op_4278_strides_0"), val = tensor([1, 1])]; - tensor var_4278_pad_0 = const()[name = string("op_4278_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4278_dilations_0 = const()[name = string("op_4278_dilations_0"), val = tensor([1, 1])]; - int32 var_4278_groups_0 = const()[name = string("op_4278_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141814848))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143387776))))[name = string("layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_4278_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4278_dilations_0, groups = var_4278_groups_0, pad = var_4278_pad_0, pad_type = var_4278_pad_type_0, strides = var_4278_strides_0, weight = layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_297_cast_fp16)[name = string("op_4278_cast_fp16")]; - string var_4284_pad_type_0 = const()[name = string("op_4284_pad_type_0"), val = string("valid")]; - tensor var_4284_strides_0 = const()[name = string("op_4284_strides_0"), val = tensor([1, 1])]; - tensor var_4284_pad_0 = const()[name = string("op_4284_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4284_dilations_0 = const()[name = string("op_4284_dilations_0"), val = tensor([1, 1])]; - int32 var_4284_groups_0 = const()[name = string("op_4284_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143436992))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143391936))))[name = string("layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4284_cast_fp16 = conv(dilations = var_4284_dilations_0, groups = var_4284_groups_0, pad = var_4284_pad_0, pad_type = var_4284_pad_type_0, strides = var_4284_strides_0, weight = layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_297_cast_fp16)[name = string("op_4284_cast_fp16")]; - tensor input_299_cast_fp16 = add(x = var_4278_cast_fp16, y = var_4284_cast_fp16)[name = string("input_299_cast_fp16")]; - tensor input_301_cast_fp16 = silu(x = input_299_cast_fp16)[name = string("input_301_cast_fp16")]; - string var_4295_pad_type_0 = const()[name = string("op_4295_pad_type_0"), val = string("valid")]; - tensor var_4295_strides_0 = const()[name = string("op_4295_strides_0"), val = tensor([1, 1])]; - tensor var_4295_pad_0 = const()[name = string("op_4295_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4295_dilations_0 = const()[name = string("op_4295_dilations_0"), val = tensor([1, 1])]; - int32 var_4295_groups_0 = const()[name = string("op_4295_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143961344))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145534272))))[name = string("layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_4295_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4295_dilations_0, groups = var_4295_groups_0, pad = var_4295_pad_0, pad_type = var_4295_pad_type_0, strides = var_4295_strides_0, weight = layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_301_cast_fp16)[name = string("op_4295_cast_fp16")]; - string var_4301_pad_type_0 = const()[name = string("op_4301_pad_type_0"), val = string("valid")]; - tensor var_4301_strides_0 = const()[name = string("op_4301_strides_0"), val = tensor([1, 1])]; - tensor var_4301_pad_0 = const()[name = string("op_4301_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4301_dilations_0 = const()[name = string("op_4301_dilations_0"), val = tensor([1, 1])]; - int32 var_4301_groups_0 = const()[name = string("op_4301_groups_0"), val = int32(1)]; - tensor layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145583680))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145535360))))[name = string("layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4301_cast_fp16 = conv(dilations = var_4301_dilations_0, groups = var_4301_groups_0, pad = var_4301_pad_0, pad_type = var_4301_pad_type_0, strides = var_4301_strides_0, weight = layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_301_cast_fp16)[name = string("op_4301_cast_fp16")]; - tensor x_67_cast_fp16 = add(x = var_4295_cast_fp16, y = var_4301_cast_fp16)[name = string("x_67_cast_fp16")]; - fp16 var_4303_to_fp16 = const()[name = string("op_4303_to_fp16"), val = fp16(0x1p-1)]; - tensor var_4304_cast_fp16 = mul(x = x_67_cast_fp16, y = var_4303_to_fp16)[name = string("op_4304_cast_fp16")]; - tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = var_4304_cast_fp16)[name = string("inputs_109_cast_fp16")]; - tensor out_109_axes_0 = const()[name = string("out_109_axes_0"), val = tensor([1])]; - fp16 var_4314_to_fp16 = const()[name = string("op_4314_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_4314_to_fp16, x = inputs_109_cast_fp16)[name = string("out_109_cast_fp16")]; - tensor inputs_111_gamma_0_to_fp16 = const()[name = string("inputs_111_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146108032)))]; - tensor inputs_111_beta_0_to_fp16 = const()[name = string("inputs_111_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146110144)))]; - fp16 inputs_111_epsilon_0_to_fp16 = const()[name = string("inputs_111_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_111_cast_fp16 = batch_norm(beta = inputs_111_beta_0_to_fp16, epsilon = inputs_111_epsilon_0_to_fp16, gamma = inputs_111_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_109_cast_fp16)[name = string("inputs_111_cast_fp16")]; - int32 var_4328 = const()[name = string("op_4328"), val = int32(3)]; - tensor out_111_axes_0 = const()[name = string("out_111_axes_0"), val = tensor([1])]; - fp16 var_4359_to_fp16 = const()[name = string("op_4359_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_4359_to_fp16, x = inputs_111_cast_fp16)[name = string("out_111_cast_fp16")]; - tensor input_303_gamma_0_to_fp16 = const()[name = string("input_303_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146112256)))]; - tensor input_303_beta_0_to_fp16 = const()[name = string("input_303_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146114368)))]; - fp16 input_303_epsilon_0_to_fp16 = const()[name = string("input_303_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_303_cast_fp16 = batch_norm(beta = input_303_beta_0_to_fp16, epsilon = input_303_epsilon_0_to_fp16, gamma = input_303_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_111_cast_fp16)[name = string("input_303_cast_fp16")]; - string var_4379_pad_type_0 = const()[name = string("op_4379_pad_type_0"), val = string("valid")]; - tensor var_4379_strides_0 = const()[name = string("op_4379_strides_0"), val = tensor([1, 1])]; - tensor var_4379_pad_0 = const()[name = string("op_4379_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4379_dilations_0 = const()[name = string("op_4379_dilations_0"), val = tensor([1, 1])]; - int32 var_4379_groups_0 = const()[name = string("op_4379_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146116480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147689408))))[name = string("layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_4379_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4379_dilations_0, groups = var_4379_groups_0, pad = var_4379_pad_0, pad_type = var_4379_pad_type_0, strides = var_4379_strides_0, weight = layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_303_cast_fp16)[name = string("op_4379_cast_fp16")]; - string var_4385_pad_type_0 = const()[name = string("op_4385_pad_type_0"), val = string("valid")]; - tensor var_4385_strides_0 = const()[name = string("op_4385_strides_0"), val = tensor([1, 1])]; - tensor var_4385_pad_0 = const()[name = string("op_4385_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4385_dilations_0 = const()[name = string("op_4385_dilations_0"), val = tensor([1, 1])]; - int32 var_4385_groups_0 = const()[name = string("op_4385_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147741184))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147693568))))[name = string("layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4385_cast_fp16 = conv(dilations = var_4385_dilations_0, groups = var_4385_groups_0, pad = var_4385_pad_0, pad_type = var_4385_pad_type_0, strides = var_4385_strides_0, weight = layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_303_cast_fp16)[name = string("op_4385_cast_fp16")]; - tensor input_305_cast_fp16 = add(x = var_4379_cast_fp16, y = var_4385_cast_fp16)[name = string("input_305_cast_fp16")]; - tensor input_307_cast_fp16 = silu(x = input_305_cast_fp16)[name = string("input_307_cast_fp16")]; - string var_4396_pad_type_0 = const()[name = string("op_4396_pad_type_0"), val = string("valid")]; - tensor var_4396_strides_0 = const()[name = string("op_4396_strides_0"), val = tensor([1, 1])]; - tensor var_4396_pad_0 = const()[name = string("op_4396_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4396_dilations_0 = const()[name = string("op_4396_dilations_0"), val = tensor([1, 1])]; - int32 var_4396_groups_0 = const()[name = string("op_4396_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148265536))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149838464))))[name = string("layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_4396_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4396_dilations_0, groups = var_4396_groups_0, pad = var_4396_pad_0, pad_type = var_4396_pad_type_0, strides = var_4396_strides_0, weight = layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_307_cast_fp16)[name = string("op_4396_cast_fp16")]; - string var_4402_pad_type_0 = const()[name = string("op_4402_pad_type_0"), val = string("valid")]; - tensor var_4402_strides_0 = const()[name = string("op_4402_strides_0"), val = tensor([1, 1])]; - tensor var_4402_pad_0 = const()[name = string("op_4402_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4402_dilations_0 = const()[name = string("op_4402_dilations_0"), val = tensor([1, 1])]; - int32 var_4402_groups_0 = const()[name = string("op_4402_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149889344))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149839552))))[name = string("layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4402_cast_fp16 = conv(dilations = var_4402_dilations_0, groups = var_4402_groups_0, pad = var_4402_pad_0, pad_type = var_4402_pad_type_0, strides = var_4402_strides_0, weight = layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_307_cast_fp16)[name = string("op_4402_cast_fp16")]; - tensor x_69_cast_fp16 = add(x = var_4396_cast_fp16, y = var_4402_cast_fp16)[name = string("x_69_cast_fp16")]; - fp16 var_4404_to_fp16 = const()[name = string("op_4404_to_fp16"), val = fp16(0x1p-1)]; - tensor var_4405_cast_fp16 = mul(x = x_69_cast_fp16, y = var_4404_to_fp16)[name = string("op_4405_cast_fp16")]; - tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = var_4405_cast_fp16)[name = string("inputs_113_cast_fp16")]; - tensor out_113_axes_0 = const()[name = string("out_113_axes_0"), val = tensor([1])]; - fp16 var_4415_to_fp16 = const()[name = string("op_4415_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_4415_to_fp16, x = inputs_113_cast_fp16)[name = string("out_113_cast_fp16")]; - tensor obj_47_gamma_0_to_fp16 = const()[name = string("obj_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150413696)))]; - tensor obj_47_beta_0_to_fp16 = const()[name = string("obj_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150415808)))]; - fp16 obj_47_epsilon_0_to_fp16 = const()[name = string("obj_47_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_47_cast_fp16 = batch_norm(beta = obj_47_beta_0_to_fp16, epsilon = obj_47_epsilon_0_to_fp16, gamma = obj_47_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_113_cast_fp16)[name = string("obj_47_cast_fp16")]; - string var_4440_pad_type_0 = const()[name = string("op_4440_pad_type_0"), val = string("valid")]; - tensor var_4440_strides_0 = const()[name = string("op_4440_strides_0"), val = tensor([1, 1])]; - tensor var_4440_pad_0 = const()[name = string("op_4440_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4440_dilations_0 = const()[name = string("op_4440_dilations_0"), val = tensor([1, 1])]; - int32 var_4440_groups_0 = const()[name = string("op_4440_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150417920))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150811200))))[name = string("layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4440_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4440_dilations_0, groups = var_4440_groups_0, pad = var_4440_pad_0, pad_type = var_4440_pad_type_0, strides = var_4440_strides_0, weight = layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = string("op_4440_cast_fp16")]; - string var_4446_pad_type_0 = const()[name = string("op_4446_pad_type_0"), val = string("valid")]; - tensor var_4446_strides_0 = const()[name = string("op_4446_strides_0"), val = tensor([1, 1])]; - tensor var_4446_pad_0 = const()[name = string("op_4446_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4446_dilations_0 = const()[name = string("op_4446_dilations_0"), val = tensor([1, 1])]; - int32 var_4446_groups_0 = const()[name = string("op_4446_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150823680))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150812288))))[name = string("layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4446_cast_fp16 = conv(dilations = var_4446_dilations_0, groups = var_4446_groups_0, pad = var_4446_pad_0, pad_type = var_4446_pad_type_0, strides = var_4446_strides_0, weight = layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = string("op_4446_cast_fp16")]; - tensor query_45_cast_fp16 = add(x = var_4440_cast_fp16, y = var_4446_cast_fp16)[name = string("query_45_cast_fp16")]; - string var_4455_pad_type_0 = const()[name = string("op_4455_pad_type_0"), val = string("valid")]; - tensor var_4455_strides_0 = const()[name = string("op_4455_strides_0"), val = tensor([1, 1])]; - tensor var_4455_pad_0 = const()[name = string("op_4455_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4455_dilations_0 = const()[name = string("op_4455_dilations_0"), val = tensor([1, 1])]; - int32 var_4455_groups_0 = const()[name = string("op_4455_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150954816))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151348096))))[name = string("layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4455_cast_fp16 = conv(dilations = var_4455_dilations_0, groups = var_4455_groups_0, pad = var_4455_pad_0, pad_type = var_4455_pad_type_0, strides = var_4455_strides_0, weight = layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = string("op_4455_cast_fp16")]; - string var_4461_pad_type_0 = const()[name = string("op_4461_pad_type_0"), val = string("valid")]; - tensor var_4461_strides_0 = const()[name = string("op_4461_strides_0"), val = tensor([1, 1])]; - tensor var_4461_pad_0 = const()[name = string("op_4461_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4461_dilations_0 = const()[name = string("op_4461_dilations_0"), val = tensor([1, 1])]; - int32 var_4461_groups_0 = const()[name = string("op_4461_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151359872))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151349184))))[name = string("layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4461_cast_fp16 = conv(dilations = var_4461_dilations_0, groups = var_4461_groups_0, pad = var_4461_pad_0, pad_type = var_4461_pad_type_0, strides = var_4461_strides_0, weight = layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = string("op_4461_cast_fp16")]; - tensor key_23_cast_fp16 = add(x = var_4455_cast_fp16, y = var_4461_cast_fp16)[name = string("key_23_cast_fp16")]; - string var_4471_pad_type_0 = const()[name = string("op_4471_pad_type_0"), val = string("valid")]; - tensor var_4471_strides_0 = const()[name = string("op_4471_strides_0"), val = tensor([1, 1])]; - tensor var_4471_pad_0 = const()[name = string("op_4471_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4471_dilations_0 = const()[name = string("op_4471_dilations_0"), val = tensor([1, 1])]; - int32 var_4471_groups_0 = const()[name = string("op_4471_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151491008))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151884288))))[name = string("layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4471_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4471_dilations_0, groups = var_4471_groups_0, pad = var_4471_pad_0, pad_type = var_4471_pad_type_0, strides = var_4471_strides_0, weight = layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = string("op_4471_cast_fp16")]; - string var_4477_pad_type_0 = const()[name = string("op_4477_pad_type_0"), val = string("valid")]; - tensor var_4477_strides_0 = const()[name = string("op_4477_strides_0"), val = tensor([1, 1])]; - tensor var_4477_pad_0 = const()[name = string("op_4477_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4477_dilations_0 = const()[name = string("op_4477_dilations_0"), val = tensor([1, 1])]; - int32 var_4477_groups_0 = const()[name = string("op_4477_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151893248))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151885376))))[name = string("layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4477_cast_fp16 = conv(dilations = var_4477_dilations_0, groups = var_4477_groups_0, pad = var_4477_pad_0, pad_type = var_4477_pad_type_0, strides = var_4477_strides_0, weight = layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = string("op_4477_cast_fp16")]; - tensor value_23_cast_fp16 = add(x = var_4471_cast_fp16, y = var_4477_cast_fp16)[name = string("value_23_cast_fp16")]; - tensor var_4480_to_fp16 = const()[name = string("op_4480_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152024384)))]; - tensor query_47_cast_fp16 = add(x = query_45_cast_fp16, y = var_4480_to_fp16)[name = string("query_47_cast_fp16")]; - tensor var_4483_to_fp16 = const()[name = string("op_4483_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152026496)))]; - tensor q_with_bias_v_23_cast_fp16 = add(x = query_45_cast_fp16, y = var_4483_to_fp16)[name = string("q_with_bias_v_23_cast_fp16")]; - string var_4493_pad_type_0 = const()[name = string("op_4493_pad_type_0"), val = string("valid")]; - tensor var_4493_strides_0 = const()[name = string("op_4493_strides_0"), val = tensor([1, 1])]; - tensor var_4493_pad_0 = const()[name = string("op_4493_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4493_dilations_0 = const()[name = string("op_4493_dilations_0"), val = tensor([1, 1])]; - int32 var_4493_groups_0 = const()[name = string("op_4493_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152028608))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152421888))))[name = string("layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_4493_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4493_dilations_0, groups = var_4493_groups_0, pad = var_4493_pad_0, pad_type = var_4493_pad_type_0, strides = var_4493_strides_0, weight = layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_4493_cast_fp16")]; - string var_4499_pad_type_0 = const()[name = string("op_4499_pad_type_0"), val = string("valid")]; - tensor var_4499_strides_0 = const()[name = string("op_4499_strides_0"), val = tensor([1, 1])]; - tensor var_4499_pad_0 = const()[name = string("op_4499_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4499_dilations_0 = const()[name = string("op_4499_dilations_0"), val = tensor([1, 1])]; - int32 var_4499_groups_0 = const()[name = string("op_4499_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152455360))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152422976))))[name = string("layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4499_cast_fp16 = conv(dilations = var_4499_dilations_0, groups = var_4499_groups_0, pad = var_4499_pad_0, pad_type = var_4499_pad_type_0, strides = var_4499_strides_0, weight = layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_4499_cast_fp16")]; - tensor p_23_cast_fp16 = add(x = var_4493_cast_fp16, y = var_4499_cast_fp16)[name = string("p_23_cast_fp16")]; - tensor var_4503 = const()[name = string("op_4503"), val = tensor([1, 8, 128, 188])]; - tensor var_4504_cast_fp16 = reshape(shape = var_4503, x = q_with_bias_v_23_cast_fp16)[name = string("op_4504_cast_fp16")]; - tensor var_4505 = const()[name = string("op_4505"), val = tensor([1, 8, 128, -1])]; - tensor var_4506_cast_fp16 = reshape(shape = var_4505, x = p_23_cast_fp16)[name = string("op_4506_cast_fp16")]; - bool matrix_bd_89_transpose_x_0 = const()[name = string("matrix_bd_89_transpose_x_0"), val = bool(true)]; - bool matrix_bd_89_transpose_y_0 = const()[name = string("matrix_bd_89_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_89_cast_fp16 = matmul(transpose_x = matrix_bd_89_transpose_x_0, transpose_y = matrix_bd_89_transpose_y_0, x = var_4504_cast_fp16, y = var_4506_cast_fp16)[name = string("matrix_bd_89_cast_fp16")]; - tensor matrix_bd_91_pad_0 = const()[name = string("matrix_bd_91_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_91_mode_0 = const()[name = string("matrix_bd_91_mode_0"), val = string("constant")]; - fp16 const_131_to_fp16 = const()[name = string("const_131_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_91_cast_fp16 = pad(constant_val = const_131_to_fp16, mode = matrix_bd_91_mode_0, pad = matrix_bd_91_pad_0, x = matrix_bd_89_cast_fp16)[name = string("matrix_bd_91_cast_fp16")]; - tensor var_4515 = const()[name = string("op_4515"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_93_cast_fp16 = reshape(shape = var_4515, x = matrix_bd_91_cast_fp16)[name = string("matrix_bd_93_cast_fp16")]; - tensor var_4519_begin_0 = const()[name = string("op_4519_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_4519_end_0 = const()[name = string("op_4519_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_4519_end_mask_0 = const()[name = string("op_4519_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_4519_cast_fp16 = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = matrix_bd_93_cast_fp16)[name = string("op_4519_cast_fp16")]; - tensor var_4520 = const()[name = string("op_4520"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_95_cast_fp16 = reshape(shape = var_4520, x = var_4519_cast_fp16)[name = string("matrix_bd_95_cast_fp16")]; - tensor var_4525_begin_0 = const()[name = string("op_4525_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4525_end_0 = const()[name = string("op_4525_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_4525_end_mask_0 = const()[name = string("op_4525_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_4525_cast_fp16 = slice_by_index(begin = var_4525_begin_0, end = var_4525_end_0, end_mask = var_4525_end_mask_0, x = matrix_bd_95_cast_fp16)[name = string("op_4525_cast_fp16")]; - fp16 var_4526_to_fp16 = const()[name = string("op_4526_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_23_cast_fp16 = mul(x = var_4525_cast_fp16, y = var_4526_to_fp16)[name = string("qk_mask_23_cast_fp16")]; - tensor var_4530 = const()[name = string("op_4530"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_23_cast_fp16 = reshape(shape = var_4530, x = query_47_cast_fp16)[name = string("mh_q_23_cast_fp16")]; - fp16 var_4532_to_fp16 = const()[name = string("op_4532_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_4533_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_4532_to_fp16)[name = string("op_4533_cast_fp16")]; - tensor var_4536 = const()[name = string("op_4536"), val = tensor([1, 8, 128, 188])]; - tensor var_4537_cast_fp16 = reshape(shape = var_4536, x = key_23_cast_fp16)[name = string("op_4537_cast_fp16")]; - bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; - bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; - tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_4533_cast_fp16, y = var_4537_cast_fp16)[name = string("mh_w_45_cast_fp16")]; - tensor mh_w_47_cast_fp16 = add(x = mh_w_45_cast_fp16, y = qk_mask_23_cast_fp16)[name = string("mh_w_47_cast_fp16")]; - tensor var_4541_cast_fp16 = softmax(axis = var_4328, x = mh_w_47_cast_fp16)[name = string("op_4541_cast_fp16")]; - tensor var_4542 = const()[name = string("op_4542"), val = tensor([1, 8, 128, 188])]; - tensor var_4543_cast_fp16 = reshape(shape = var_4542, x = value_23_cast_fp16)[name = string("op_4543_cast_fp16")]; - bool attn_23_transpose_x_0 = const()[name = string("attn_23_transpose_x_0"), val = bool(false)]; - bool attn_23_transpose_y_0 = const()[name = string("attn_23_transpose_y_0"), val = bool(true)]; - tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_4543_cast_fp16, y = var_4541_cast_fp16)[name = string("attn_23_cast_fp16")]; - tensor var_4546 = const()[name = string("op_4546"), val = tensor([1, 1024, 1, 188])]; - tensor input_309_cast_fp16 = reshape(shape = var_4546, x = attn_23_cast_fp16)[name = string("input_309_cast_fp16")]; - string var_4556_pad_type_0 = const()[name = string("op_4556_pad_type_0"), val = string("valid")]; - tensor var_4556_strides_0 = const()[name = string("op_4556_strides_0"), val = tensor([1, 1])]; - tensor var_4556_pad_0 = const()[name = string("op_4556_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4556_dilations_0 = const()[name = string("op_4556_dilations_0"), val = tensor([1, 1])]; - int32 var_4556_groups_0 = const()[name = string("op_4556_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152586496))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152979776))))[name = string("layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4556_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4556_dilations_0, groups = var_4556_groups_0, pad = var_4556_pad_0, pad_type = var_4556_pad_type_0, strides = var_4556_strides_0, weight = layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_309_cast_fp16)[name = string("op_4556_cast_fp16")]; - string var_4562_pad_type_0 = const()[name = string("op_4562_pad_type_0"), val = string("valid")]; - tensor var_4562_strides_0 = const()[name = string("op_4562_strides_0"), val = tensor([1, 1])]; - tensor var_4562_pad_0 = const()[name = string("op_4562_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4562_dilations_0 = const()[name = string("op_4562_dilations_0"), val = tensor([1, 1])]; - int32 var_4562_groups_0 = const()[name = string("op_4562_groups_0"), val = int32(1)]; - tensor layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152989376))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152980864))))[name = string("layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4562_cast_fp16 = conv(dilations = var_4562_dilations_0, groups = var_4562_groups_0, pad = var_4562_pad_0, pad_type = var_4562_pad_type_0, strides = var_4562_strides_0, weight = layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_309_cast_fp16)[name = string("op_4562_cast_fp16")]; - tensor obj_49_cast_fp16 = add(x = var_4556_cast_fp16, y = var_4562_cast_fp16)[name = string("obj_49_cast_fp16")]; - tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_115_cast_fp16")]; - tensor out_115_axes_0 = const()[name = string("out_115_axes_0"), val = tensor([1])]; - fp16 var_4573_to_fp16 = const()[name = string("op_4573_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_4573_to_fp16, x = inputs_115_cast_fp16)[name = string("out_115_cast_fp16")]; - tensor input_311_gamma_0_to_fp16 = const()[name = string("input_311_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153120512)))]; - tensor input_311_beta_0_to_fp16 = const()[name = string("input_311_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153122624)))]; - fp16 input_311_epsilon_0_to_fp16 = const()[name = string("input_311_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_311_cast_fp16 = batch_norm(beta = input_311_beta_0_to_fp16, epsilon = input_311_epsilon_0_to_fp16, gamma = input_311_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_115_cast_fp16)[name = string("input_311_cast_fp16")]; - string var_4594_pad_type_0 = const()[name = string("op_4594_pad_type_0"), val = string("valid")]; - tensor var_4594_strides_0 = const()[name = string("op_4594_strides_0"), val = tensor([1, 1])]; - tensor var_4594_pad_0 = const()[name = string("op_4594_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4594_dilations_0 = const()[name = string("op_4594_dilations_0"), val = tensor([1, 1])]; - int32 var_4594_groups_0 = const()[name = string("op_4594_groups_0"), val = int32(1)]; - tensor layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153124736))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153911232))))[name = string("layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_4594_cast_fp16 = conv(dilations = var_4594_dilations_0, groups = var_4594_groups_0, pad = var_4594_pad_0, pad_type = var_4594_pad_type_0, strides = var_4594_strides_0, weight = layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_311_cast_fp16)[name = string("op_4594_cast_fp16")]; - string var_4600_pad_type_0 = const()[name = string("op_4600_pad_type_0"), val = string("valid")]; - tensor var_4600_strides_0 = const()[name = string("op_4600_strides_0"), val = tensor([1, 1])]; - tensor var_4600_pad_0 = const()[name = string("op_4600_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4600_dilations_0 = const()[name = string("op_4600_dilations_0"), val = tensor([1, 1])]; - int32 var_4600_groups_0 = const()[name = string("op_4600_groups_0"), val = int32(1)]; - tensor layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153931520))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153913344))))[name = string("layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4600_cast_fp16 = conv(dilations = var_4600_dilations_0, groups = var_4600_groups_0, pad = var_4600_pad_0, pad_type = var_4600_pad_type_0, strides = var_4600_strides_0, weight = layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_311_cast_fp16)[name = string("op_4600_cast_fp16")]; - tensor input_313_cast_fp16 = add(x = var_4594_cast_fp16, y = var_4600_cast_fp16)[name = string("input_313_cast_fp16")]; - int32 input_315_split_num_splits_0 = const()[name = string("input_315_split_num_splits_0"), val = int32(2)]; - int32 input_315_split_axis_0 = const()[name = string("input_315_split_axis_0"), val = int32(1)]; - tensor input_315_split_cast_fp16_0, tensor input_315_split_cast_fp16_1 = split(axis = input_315_split_axis_0, num_splits = input_315_split_num_splits_0, x = input_313_cast_fp16)[name = string("input_315_split_cast_fp16")]; - tensor input_315_split_1_sigmoid_cast_fp16 = sigmoid(x = input_315_split_cast_fp16_1)[name = string("input_315_split_1_sigmoid_cast_fp16")]; - tensor input_315_cast_fp16 = mul(x = input_315_split_cast_fp16_0, y = input_315_split_1_sigmoid_cast_fp16)[name = string("input_315_cast_fp16")]; - string input_317_pad_type_0 = const()[name = string("input_317_pad_type_0"), val = string("custom")]; - tensor input_317_pad_0 = const()[name = string("input_317_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_317_groups_0 = const()[name = string("input_317_groups_0"), val = int32(1024)]; - tensor input_317_strides_0 = const()[name = string("input_317_strides_0"), val = tensor([1, 1])]; - tensor input_317_dilations_0 = const()[name = string("input_317_dilations_0"), val = tensor([1, 1])]; - tensor const_290_to_fp16 = const()[name = string("const_290_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154193728)))]; - tensor const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154212224)))]; - tensor input_319_cast_fp16 = conv(bias = const_291_to_fp16, dilations = input_317_dilations_0, groups = input_317_groups_0, pad = input_317_pad_0, pad_type = input_317_pad_type_0, strides = input_317_strides_0, weight = const_290_to_fp16, x = input_315_cast_fp16)[name = string("input_319_cast_fp16")]; - tensor input_321_cast_fp16 = silu(x = input_319_cast_fp16)[name = string("input_321_cast_fp16")]; - string var_4622_pad_type_0 = const()[name = string("op_4622_pad_type_0"), val = string("valid")]; - tensor var_4622_strides_0 = const()[name = string("op_4622_strides_0"), val = tensor([1, 1])]; - tensor var_4622_pad_0 = const()[name = string("op_4622_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4622_dilations_0 = const()[name = string("op_4622_dilations_0"), val = tensor([1, 1])]; - int32 var_4622_groups_0 = const()[name = string("op_4622_groups_0"), val = int32(1)]; - tensor layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154214336))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154607616))))[name = string("layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_4622_cast_fp16 = conv(dilations = var_4622_dilations_0, groups = var_4622_groups_0, pad = var_4622_pad_0, pad_type = var_4622_pad_type_0, strides = var_4622_strides_0, weight = layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_321_cast_fp16)[name = string("op_4622_cast_fp16")]; - string var_4628_pad_type_0 = const()[name = string("op_4628_pad_type_0"), val = string("valid")]; - tensor var_4628_strides_0 = const()[name = string("op_4628_strides_0"), val = tensor([1, 1])]; - tensor var_4628_pad_0 = const()[name = string("op_4628_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4628_dilations_0 = const()[name = string("op_4628_dilations_0"), val = tensor([1, 1])]; - int32 var_4628_groups_0 = const()[name = string("op_4628_groups_0"), val = int32(1)]; - tensor layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154618112))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154608704))))[name = string("layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4628_cast_fp16 = conv(dilations = var_4628_dilations_0, groups = var_4628_groups_0, pad = var_4628_pad_0, pad_type = var_4628_pad_type_0, strides = var_4628_strides_0, weight = layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_321_cast_fp16)[name = string("op_4628_cast_fp16")]; - tensor x_71_cast_fp16 = add(x = var_4622_cast_fp16, y = var_4628_cast_fp16)[name = string("x_71_cast_fp16")]; - tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = x_71_cast_fp16)[name = string("inputs_117_cast_fp16")]; - tensor out_117_axes_0 = const()[name = string("out_117_axes_0"), val = tensor([1])]; - fp16 var_4639_to_fp16 = const()[name = string("op_4639_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_4639_to_fp16, x = inputs_117_cast_fp16)[name = string("out_117_cast_fp16")]; - tensor input_323_gamma_0_to_fp16 = const()[name = string("input_323_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154749248)))]; - tensor input_323_beta_0_to_fp16 = const()[name = string("input_323_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154751360)))]; - fp16 input_323_epsilon_0_to_fp16 = const()[name = string("input_323_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_323_cast_fp16 = batch_norm(beta = input_323_beta_0_to_fp16, epsilon = input_323_epsilon_0_to_fp16, gamma = input_323_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_117_cast_fp16)[name = string("input_323_cast_fp16")]; - string var_4659_pad_type_0 = const()[name = string("op_4659_pad_type_0"), val = string("valid")]; - tensor var_4659_strides_0 = const()[name = string("op_4659_strides_0"), val = tensor([1, 1])]; - tensor var_4659_pad_0 = const()[name = string("op_4659_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4659_dilations_0 = const()[name = string("op_4659_dilations_0"), val = tensor([1, 1])]; - int32 var_4659_groups_0 = const()[name = string("op_4659_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154753472))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156326400))))[name = string("layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_4659_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4659_dilations_0, groups = var_4659_groups_0, pad = var_4659_pad_0, pad_type = var_4659_pad_type_0, strides = var_4659_strides_0, weight = layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_323_cast_fp16)[name = string("op_4659_cast_fp16")]; - string var_4665_pad_type_0 = const()[name = string("op_4665_pad_type_0"), val = string("valid")]; - tensor var_4665_strides_0 = const()[name = string("op_4665_strides_0"), val = tensor([1, 1])]; - tensor var_4665_pad_0 = const()[name = string("op_4665_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4665_dilations_0 = const()[name = string("op_4665_dilations_0"), val = tensor([1, 1])]; - int32 var_4665_groups_0 = const()[name = string("op_4665_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156377280))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156330560))))[name = string("layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4665_cast_fp16 = conv(dilations = var_4665_dilations_0, groups = var_4665_groups_0, pad = var_4665_pad_0, pad_type = var_4665_pad_type_0, strides = var_4665_strides_0, weight = layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_323_cast_fp16)[name = string("op_4665_cast_fp16")]; - tensor input_325_cast_fp16 = add(x = var_4659_cast_fp16, y = var_4665_cast_fp16)[name = string("input_325_cast_fp16")]; - tensor input_327_cast_fp16 = silu(x = input_325_cast_fp16)[name = string("input_327_cast_fp16")]; - string var_4676_pad_type_0 = const()[name = string("op_4676_pad_type_0"), val = string("valid")]; - tensor var_4676_strides_0 = const()[name = string("op_4676_strides_0"), val = tensor([1, 1])]; - tensor var_4676_pad_0 = const()[name = string("op_4676_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4676_dilations_0 = const()[name = string("op_4676_dilations_0"), val = tensor([1, 1])]; - int32 var_4676_groups_0 = const()[name = string("op_4676_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156901632))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158474560))))[name = string("layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_4676_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4676_dilations_0, groups = var_4676_groups_0, pad = var_4676_pad_0, pad_type = var_4676_pad_type_0, strides = var_4676_strides_0, weight = layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_327_cast_fp16)[name = string("op_4676_cast_fp16")]; - string var_4682_pad_type_0 = const()[name = string("op_4682_pad_type_0"), val = string("valid")]; - tensor var_4682_strides_0 = const()[name = string("op_4682_strides_0"), val = tensor([1, 1])]; - tensor var_4682_pad_0 = const()[name = string("op_4682_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4682_dilations_0 = const()[name = string("op_4682_dilations_0"), val = tensor([1, 1])]; - int32 var_4682_groups_0 = const()[name = string("op_4682_groups_0"), val = int32(1)]; - tensor layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158530240))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158475648))))[name = string("layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4682_cast_fp16 = conv(dilations = var_4682_dilations_0, groups = var_4682_groups_0, pad = var_4682_pad_0, pad_type = var_4682_pad_type_0, strides = var_4682_strides_0, weight = layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_327_cast_fp16)[name = string("op_4682_cast_fp16")]; - tensor x_73_cast_fp16 = add(x = var_4676_cast_fp16, y = var_4682_cast_fp16)[name = string("x_73_cast_fp16")]; - fp16 var_4684_to_fp16 = const()[name = string("op_4684_to_fp16"), val = fp16(0x1p-1)]; - tensor var_4685_cast_fp16 = mul(x = x_73_cast_fp16, y = var_4684_to_fp16)[name = string("op_4685_cast_fp16")]; - tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = var_4685_cast_fp16)[name = string("inputs_119_cast_fp16")]; - tensor out_119_axes_0 = const()[name = string("out_119_axes_0"), val = tensor([1])]; - fp16 var_4695_to_fp16 = const()[name = string("op_4695_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_4695_to_fp16, x = inputs_119_cast_fp16)[name = string("out_119_cast_fp16")]; - tensor inputs_121_gamma_0_to_fp16 = const()[name = string("inputs_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159054592)))]; - tensor inputs_121_beta_0_to_fp16 = const()[name = string("inputs_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159056704)))]; - fp16 inputs_121_epsilon_0_to_fp16 = const()[name = string("inputs_121_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_121_cast_fp16 = batch_norm(beta = inputs_121_beta_0_to_fp16, epsilon = inputs_121_epsilon_0_to_fp16, gamma = inputs_121_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_119_cast_fp16)[name = string("inputs_121_cast_fp16")]; - int32 var_4709 = const()[name = string("op_4709"), val = int32(3)]; - tensor out_121_axes_0 = const()[name = string("out_121_axes_0"), val = tensor([1])]; - fp16 var_4740_to_fp16 = const()[name = string("op_4740_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_4740_to_fp16, x = inputs_121_cast_fp16)[name = string("out_121_cast_fp16")]; - tensor input_329_gamma_0_to_fp16 = const()[name = string("input_329_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159058816)))]; - tensor input_329_beta_0_to_fp16 = const()[name = string("input_329_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159060928)))]; - fp16 input_329_epsilon_0_to_fp16 = const()[name = string("input_329_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_329_cast_fp16 = batch_norm(beta = input_329_beta_0_to_fp16, epsilon = input_329_epsilon_0_to_fp16, gamma = input_329_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_121_cast_fp16)[name = string("input_329_cast_fp16")]; - string var_4760_pad_type_0 = const()[name = string("op_4760_pad_type_0"), val = string("valid")]; - tensor var_4760_strides_0 = const()[name = string("op_4760_strides_0"), val = tensor([1, 1])]; - tensor var_4760_pad_0 = const()[name = string("op_4760_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4760_dilations_0 = const()[name = string("op_4760_dilations_0"), val = tensor([1, 1])]; - int32 var_4760_groups_0 = const()[name = string("op_4760_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159063040))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160635968))))[name = string("layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_4760_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4760_dilations_0, groups = var_4760_groups_0, pad = var_4760_pad_0, pad_type = var_4760_pad_type_0, strides = var_4760_strides_0, weight = layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_329_cast_fp16)[name = string("op_4760_cast_fp16")]; - string var_4766_pad_type_0 = const()[name = string("op_4766_pad_type_0"), val = string("valid")]; - tensor var_4766_strides_0 = const()[name = string("op_4766_strides_0"), val = tensor([1, 1])]; - tensor var_4766_pad_0 = const()[name = string("op_4766_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4766_dilations_0 = const()[name = string("op_4766_dilations_0"), val = tensor([1, 1])]; - int32 var_4766_groups_0 = const()[name = string("op_4766_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160689856))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160640128))))[name = string("layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4766_cast_fp16 = conv(dilations = var_4766_dilations_0, groups = var_4766_groups_0, pad = var_4766_pad_0, pad_type = var_4766_pad_type_0, strides = var_4766_strides_0, weight = layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_329_cast_fp16)[name = string("op_4766_cast_fp16")]; - tensor input_331_cast_fp16 = add(x = var_4760_cast_fp16, y = var_4766_cast_fp16)[name = string("input_331_cast_fp16")]; - tensor input_333_cast_fp16 = silu(x = input_331_cast_fp16)[name = string("input_333_cast_fp16")]; - string var_4777_pad_type_0 = const()[name = string("op_4777_pad_type_0"), val = string("valid")]; - tensor var_4777_strides_0 = const()[name = string("op_4777_strides_0"), val = tensor([1, 1])]; - tensor var_4777_pad_0 = const()[name = string("op_4777_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4777_dilations_0 = const()[name = string("op_4777_dilations_0"), val = tensor([1, 1])]; - int32 var_4777_groups_0 = const()[name = string("op_4777_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161214208))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162787136))))[name = string("layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_4777_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4777_dilations_0, groups = var_4777_groups_0, pad = var_4777_pad_0, pad_type = var_4777_pad_type_0, strides = var_4777_strides_0, weight = layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_333_cast_fp16)[name = string("op_4777_cast_fp16")]; - string var_4783_pad_type_0 = const()[name = string("op_4783_pad_type_0"), val = string("valid")]; - tensor var_4783_strides_0 = const()[name = string("op_4783_strides_0"), val = tensor([1, 1])]; - tensor var_4783_pad_0 = const()[name = string("op_4783_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4783_dilations_0 = const()[name = string("op_4783_dilations_0"), val = tensor([1, 1])]; - int32 var_4783_groups_0 = const()[name = string("op_4783_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162841280))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162788224))))[name = string("layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4783_cast_fp16 = conv(dilations = var_4783_dilations_0, groups = var_4783_groups_0, pad = var_4783_pad_0, pad_type = var_4783_pad_type_0, strides = var_4783_strides_0, weight = layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_333_cast_fp16)[name = string("op_4783_cast_fp16")]; - tensor x_75_cast_fp16 = add(x = var_4777_cast_fp16, y = var_4783_cast_fp16)[name = string("x_75_cast_fp16")]; - fp16 var_4785_to_fp16 = const()[name = string("op_4785_to_fp16"), val = fp16(0x1p-1)]; - tensor var_4786_cast_fp16 = mul(x = x_75_cast_fp16, y = var_4785_to_fp16)[name = string("op_4786_cast_fp16")]; - tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = var_4786_cast_fp16)[name = string("inputs_123_cast_fp16")]; - tensor out_123_axes_0 = const()[name = string("out_123_axes_0"), val = tensor([1])]; - fp16 var_4796_to_fp16 = const()[name = string("op_4796_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_4796_to_fp16, x = inputs_123_cast_fp16)[name = string("out_123_cast_fp16")]; - tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163365632)))]; - tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163367744)))]; - fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_123_cast_fp16)[name = string("obj_51_cast_fp16")]; - string var_4821_pad_type_0 = const()[name = string("op_4821_pad_type_0"), val = string("valid")]; - tensor var_4821_strides_0 = const()[name = string("op_4821_strides_0"), val = tensor([1, 1])]; - tensor var_4821_pad_0 = const()[name = string("op_4821_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4821_dilations_0 = const()[name = string("op_4821_dilations_0"), val = tensor([1, 1])]; - int32 var_4821_groups_0 = const()[name = string("op_4821_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163369856))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163763136))))[name = string("layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4821_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4821_dilations_0, groups = var_4821_groups_0, pad = var_4821_pad_0, pad_type = var_4821_pad_type_0, strides = var_4821_strides_0, weight = layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = string("op_4821_cast_fp16")]; - string var_4827_pad_type_0 = const()[name = string("op_4827_pad_type_0"), val = string("valid")]; - tensor var_4827_strides_0 = const()[name = string("op_4827_strides_0"), val = tensor([1, 1])]; - tensor var_4827_pad_0 = const()[name = string("op_4827_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4827_dilations_0 = const()[name = string("op_4827_dilations_0"), val = tensor([1, 1])]; - int32 var_4827_groups_0 = const()[name = string("op_4827_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163775232))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163764224))))[name = string("layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4827_cast_fp16 = conv(dilations = var_4827_dilations_0, groups = var_4827_groups_0, pad = var_4827_pad_0, pad_type = var_4827_pad_type_0, strides = var_4827_strides_0, weight = layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = string("op_4827_cast_fp16")]; - tensor query_49_cast_fp16 = add(x = var_4821_cast_fp16, y = var_4827_cast_fp16)[name = string("query_49_cast_fp16")]; - string var_4836_pad_type_0 = const()[name = string("op_4836_pad_type_0"), val = string("valid")]; - tensor var_4836_strides_0 = const()[name = string("op_4836_strides_0"), val = tensor([1, 1])]; - tensor var_4836_pad_0 = const()[name = string("op_4836_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4836_dilations_0 = const()[name = string("op_4836_dilations_0"), val = tensor([1, 1])]; - int32 var_4836_groups_0 = const()[name = string("op_4836_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163906368))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164299648))))[name = string("layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4836_cast_fp16 = conv(dilations = var_4836_dilations_0, groups = var_4836_groups_0, pad = var_4836_pad_0, pad_type = var_4836_pad_type_0, strides = var_4836_strides_0, weight = layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = string("op_4836_cast_fp16")]; - string var_4842_pad_type_0 = const()[name = string("op_4842_pad_type_0"), val = string("valid")]; - tensor var_4842_strides_0 = const()[name = string("op_4842_strides_0"), val = tensor([1, 1])]; - tensor var_4842_pad_0 = const()[name = string("op_4842_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4842_dilations_0 = const()[name = string("op_4842_dilations_0"), val = tensor([1, 1])]; - int32 var_4842_groups_0 = const()[name = string("op_4842_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164310784))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164300736))))[name = string("layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4842_cast_fp16 = conv(dilations = var_4842_dilations_0, groups = var_4842_groups_0, pad = var_4842_pad_0, pad_type = var_4842_pad_type_0, strides = var_4842_strides_0, weight = layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = string("op_4842_cast_fp16")]; - tensor key_25_cast_fp16 = add(x = var_4836_cast_fp16, y = var_4842_cast_fp16)[name = string("key_25_cast_fp16")]; - string var_4852_pad_type_0 = const()[name = string("op_4852_pad_type_0"), val = string("valid")]; - tensor var_4852_strides_0 = const()[name = string("op_4852_strides_0"), val = tensor([1, 1])]; - tensor var_4852_pad_0 = const()[name = string("op_4852_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4852_dilations_0 = const()[name = string("op_4852_dilations_0"), val = tensor([1, 1])]; - int32 var_4852_groups_0 = const()[name = string("op_4852_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164441920))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164835200))))[name = string("layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4852_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4852_dilations_0, groups = var_4852_groups_0, pad = var_4852_pad_0, pad_type = var_4852_pad_type_0, strides = var_4852_strides_0, weight = layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = string("op_4852_cast_fp16")]; - string var_4858_pad_type_0 = const()[name = string("op_4858_pad_type_0"), val = string("valid")]; - tensor var_4858_strides_0 = const()[name = string("op_4858_strides_0"), val = tensor([1, 1])]; - tensor var_4858_pad_0 = const()[name = string("op_4858_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4858_dilations_0 = const()[name = string("op_4858_dilations_0"), val = tensor([1, 1])]; - int32 var_4858_groups_0 = const()[name = string("op_4858_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164844544))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164836288))))[name = string("layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4858_cast_fp16 = conv(dilations = var_4858_dilations_0, groups = var_4858_groups_0, pad = var_4858_pad_0, pad_type = var_4858_pad_type_0, strides = var_4858_strides_0, weight = layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = string("op_4858_cast_fp16")]; - tensor value_25_cast_fp16 = add(x = var_4852_cast_fp16, y = var_4858_cast_fp16)[name = string("value_25_cast_fp16")]; - tensor var_4861_to_fp16 = const()[name = string("op_4861_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164975680)))]; - tensor query_51_cast_fp16 = add(x = query_49_cast_fp16, y = var_4861_to_fp16)[name = string("query_51_cast_fp16")]; - tensor var_4864_to_fp16 = const()[name = string("op_4864_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164977792)))]; - tensor q_with_bias_v_25_cast_fp16 = add(x = query_49_cast_fp16, y = var_4864_to_fp16)[name = string("q_with_bias_v_25_cast_fp16")]; - string var_4874_pad_type_0 = const()[name = string("op_4874_pad_type_0"), val = string("valid")]; - tensor var_4874_strides_0 = const()[name = string("op_4874_strides_0"), val = tensor([1, 1])]; - tensor var_4874_pad_0 = const()[name = string("op_4874_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4874_dilations_0 = const()[name = string("op_4874_dilations_0"), val = tensor([1, 1])]; - int32 var_4874_groups_0 = const()[name = string("op_4874_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164979904))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165373184))))[name = string("layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_4874_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4874_dilations_0, groups = var_4874_groups_0, pad = var_4874_pad_0, pad_type = var_4874_pad_type_0, strides = var_4874_strides_0, weight = layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_4874_cast_fp16")]; - string var_4880_pad_type_0 = const()[name = string("op_4880_pad_type_0"), val = string("valid")]; - tensor var_4880_strides_0 = const()[name = string("op_4880_strides_0"), val = tensor([1, 1])]; - tensor var_4880_pad_0 = const()[name = string("op_4880_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4880_dilations_0 = const()[name = string("op_4880_dilations_0"), val = tensor([1, 1])]; - int32 var_4880_groups_0 = const()[name = string("op_4880_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165409920))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165374272))))[name = string("layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4880_cast_fp16 = conv(dilations = var_4880_dilations_0, groups = var_4880_groups_0, pad = var_4880_pad_0, pad_type = var_4880_pad_type_0, strides = var_4880_strides_0, weight = layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_4880_cast_fp16")]; - tensor p_25_cast_fp16 = add(x = var_4874_cast_fp16, y = var_4880_cast_fp16)[name = string("p_25_cast_fp16")]; - tensor var_4884 = const()[name = string("op_4884"), val = tensor([1, 8, 128, 188])]; - tensor var_4885_cast_fp16 = reshape(shape = var_4884, x = q_with_bias_v_25_cast_fp16)[name = string("op_4885_cast_fp16")]; - tensor var_4886 = const()[name = string("op_4886"), val = tensor([1, 8, 128, -1])]; - tensor var_4887_cast_fp16 = reshape(shape = var_4886, x = p_25_cast_fp16)[name = string("op_4887_cast_fp16")]; - bool matrix_bd_97_transpose_x_0 = const()[name = string("matrix_bd_97_transpose_x_0"), val = bool(true)]; - bool matrix_bd_97_transpose_y_0 = const()[name = string("matrix_bd_97_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_97_cast_fp16 = matmul(transpose_x = matrix_bd_97_transpose_x_0, transpose_y = matrix_bd_97_transpose_y_0, x = var_4885_cast_fp16, y = var_4887_cast_fp16)[name = string("matrix_bd_97_cast_fp16")]; - tensor matrix_bd_99_pad_0 = const()[name = string("matrix_bd_99_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_99_mode_0 = const()[name = string("matrix_bd_99_mode_0"), val = string("constant")]; - fp16 const_142_to_fp16 = const()[name = string("const_142_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_99_cast_fp16 = pad(constant_val = const_142_to_fp16, mode = matrix_bd_99_mode_0, pad = matrix_bd_99_pad_0, x = matrix_bd_97_cast_fp16)[name = string("matrix_bd_99_cast_fp16")]; - tensor var_4896 = const()[name = string("op_4896"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_101_cast_fp16 = reshape(shape = var_4896, x = matrix_bd_99_cast_fp16)[name = string("matrix_bd_101_cast_fp16")]; - tensor var_4900_begin_0 = const()[name = string("op_4900_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_4900_end_0 = const()[name = string("op_4900_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_4900_end_mask_0 = const()[name = string("op_4900_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_4900_cast_fp16 = slice_by_index(begin = var_4900_begin_0, end = var_4900_end_0, end_mask = var_4900_end_mask_0, x = matrix_bd_101_cast_fp16)[name = string("op_4900_cast_fp16")]; - tensor var_4901 = const()[name = string("op_4901"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_103_cast_fp16 = reshape(shape = var_4901, x = var_4900_cast_fp16)[name = string("matrix_bd_103_cast_fp16")]; - tensor var_4906_begin_0 = const()[name = string("op_4906_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4906_end_0 = const()[name = string("op_4906_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_4906_end_mask_0 = const()[name = string("op_4906_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_4906_cast_fp16 = slice_by_index(begin = var_4906_begin_0, end = var_4906_end_0, end_mask = var_4906_end_mask_0, x = matrix_bd_103_cast_fp16)[name = string("op_4906_cast_fp16")]; - fp16 var_4907_to_fp16 = const()[name = string("op_4907_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_25_cast_fp16 = mul(x = var_4906_cast_fp16, y = var_4907_to_fp16)[name = string("qk_mask_25_cast_fp16")]; - tensor var_4911 = const()[name = string("op_4911"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_25_cast_fp16 = reshape(shape = var_4911, x = query_51_cast_fp16)[name = string("mh_q_25_cast_fp16")]; - fp16 var_4913_to_fp16 = const()[name = string("op_4913_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_4914_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_4913_to_fp16)[name = string("op_4914_cast_fp16")]; - tensor var_4917 = const()[name = string("op_4917"), val = tensor([1, 8, 128, 188])]; - tensor var_4918_cast_fp16 = reshape(shape = var_4917, x = key_25_cast_fp16)[name = string("op_4918_cast_fp16")]; - bool mh_w_49_transpose_x_0 = const()[name = string("mh_w_49_transpose_x_0"), val = bool(true)]; - bool mh_w_49_transpose_y_0 = const()[name = string("mh_w_49_transpose_y_0"), val = bool(false)]; - tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_4914_cast_fp16, y = var_4918_cast_fp16)[name = string("mh_w_49_cast_fp16")]; - tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = qk_mask_25_cast_fp16)[name = string("mh_w_51_cast_fp16")]; - tensor var_4922_cast_fp16 = softmax(axis = var_4709, x = mh_w_51_cast_fp16)[name = string("op_4922_cast_fp16")]; - tensor var_4923 = const()[name = string("op_4923"), val = tensor([1, 8, 128, 188])]; - tensor var_4924_cast_fp16 = reshape(shape = var_4923, x = value_25_cast_fp16)[name = string("op_4924_cast_fp16")]; - bool attn_25_transpose_x_0 = const()[name = string("attn_25_transpose_x_0"), val = bool(false)]; - bool attn_25_transpose_y_0 = const()[name = string("attn_25_transpose_y_0"), val = bool(true)]; - tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_4924_cast_fp16, y = var_4922_cast_fp16)[name = string("attn_25_cast_fp16")]; - tensor var_4927 = const()[name = string("op_4927"), val = tensor([1, 1024, 1, 188])]; - tensor input_335_cast_fp16 = reshape(shape = var_4927, x = attn_25_cast_fp16)[name = string("input_335_cast_fp16")]; - string var_4937_pad_type_0 = const()[name = string("op_4937_pad_type_0"), val = string("valid")]; - tensor var_4937_strides_0 = const()[name = string("op_4937_strides_0"), val = tensor([1, 1])]; - tensor var_4937_pad_0 = const()[name = string("op_4937_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4937_dilations_0 = const()[name = string("op_4937_dilations_0"), val = tensor([1, 1])]; - int32 var_4937_groups_0 = const()[name = string("op_4937_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165541056))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165934336))))[name = string("layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_4937_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4937_dilations_0, groups = var_4937_groups_0, pad = var_4937_pad_0, pad_type = var_4937_pad_type_0, strides = var_4937_strides_0, weight = layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_335_cast_fp16)[name = string("op_4937_cast_fp16")]; - string var_4943_pad_type_0 = const()[name = string("op_4943_pad_type_0"), val = string("valid")]; - tensor var_4943_strides_0 = const()[name = string("op_4943_strides_0"), val = tensor([1, 1])]; - tensor var_4943_pad_0 = const()[name = string("op_4943_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4943_dilations_0 = const()[name = string("op_4943_dilations_0"), val = tensor([1, 1])]; - int32 var_4943_groups_0 = const()[name = string("op_4943_groups_0"), val = int32(1)]; - tensor layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165944384))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165935424))))[name = string("layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4943_cast_fp16 = conv(dilations = var_4943_dilations_0, groups = var_4943_groups_0, pad = var_4943_pad_0, pad_type = var_4943_pad_type_0, strides = var_4943_strides_0, weight = layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_335_cast_fp16)[name = string("op_4943_cast_fp16")]; - tensor obj_53_cast_fp16 = add(x = var_4937_cast_fp16, y = var_4943_cast_fp16)[name = string("obj_53_cast_fp16")]; - tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = obj_53_cast_fp16)[name = string("inputs_125_cast_fp16")]; - tensor out_125_axes_0 = const()[name = string("out_125_axes_0"), val = tensor([1])]; - fp16 var_4954_to_fp16 = const()[name = string("op_4954_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_4954_to_fp16, x = inputs_125_cast_fp16)[name = string("out_125_cast_fp16")]; - tensor input_337_gamma_0_to_fp16 = const()[name = string("input_337_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166075520)))]; - tensor input_337_beta_0_to_fp16 = const()[name = string("input_337_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166077632)))]; - fp16 input_337_epsilon_0_to_fp16 = const()[name = string("input_337_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_337_cast_fp16 = batch_norm(beta = input_337_beta_0_to_fp16, epsilon = input_337_epsilon_0_to_fp16, gamma = input_337_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_125_cast_fp16)[name = string("input_337_cast_fp16")]; - string var_4975_pad_type_0 = const()[name = string("op_4975_pad_type_0"), val = string("valid")]; - tensor var_4975_strides_0 = const()[name = string("op_4975_strides_0"), val = tensor([1, 1])]; - tensor var_4975_pad_0 = const()[name = string("op_4975_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4975_dilations_0 = const()[name = string("op_4975_dilations_0"), val = tensor([1, 1])]; - int32 var_4975_groups_0 = const()[name = string("op_4975_groups_0"), val = int32(1)]; - tensor layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166079744))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166866240))))[name = string("layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_4975_cast_fp16 = conv(dilations = var_4975_dilations_0, groups = var_4975_groups_0, pad = var_4975_pad_0, pad_type = var_4975_pad_type_0, strides = var_4975_strides_0, weight = layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = string("op_4975_cast_fp16")]; - string var_4981_pad_type_0 = const()[name = string("op_4981_pad_type_0"), val = string("valid")]; - tensor var_4981_strides_0 = const()[name = string("op_4981_strides_0"), val = tensor([1, 1])]; - tensor var_4981_pad_0 = const()[name = string("op_4981_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_4981_dilations_0 = const()[name = string("op_4981_dilations_0"), val = tensor([1, 1])]; - int32 var_4981_groups_0 = const()[name = string("op_4981_groups_0"), val = int32(1)]; - tensor layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166888512))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166868352))))[name = string("layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_4981_cast_fp16 = conv(dilations = var_4981_dilations_0, groups = var_4981_groups_0, pad = var_4981_pad_0, pad_type = var_4981_pad_type_0, strides = var_4981_strides_0, weight = layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_337_cast_fp16)[name = string("op_4981_cast_fp16")]; - tensor input_339_cast_fp16 = add(x = var_4975_cast_fp16, y = var_4981_cast_fp16)[name = string("input_339_cast_fp16")]; - int32 input_341_split_num_splits_0 = const()[name = string("input_341_split_num_splits_0"), val = int32(2)]; - int32 input_341_split_axis_0 = const()[name = string("input_341_split_axis_0"), val = int32(1)]; - tensor input_341_split_cast_fp16_0, tensor input_341_split_cast_fp16_1 = split(axis = input_341_split_axis_0, num_splits = input_341_split_num_splits_0, x = input_339_cast_fp16)[name = string("input_341_split_cast_fp16")]; - tensor input_341_split_1_sigmoid_cast_fp16 = sigmoid(x = input_341_split_cast_fp16_1)[name = string("input_341_split_1_sigmoid_cast_fp16")]; - tensor input_341_cast_fp16 = mul(x = input_341_split_cast_fp16_0, y = input_341_split_1_sigmoid_cast_fp16)[name = string("input_341_cast_fp16")]; - string input_343_pad_type_0 = const()[name = string("input_343_pad_type_0"), val = string("custom")]; - tensor input_343_pad_0 = const()[name = string("input_343_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_343_groups_0 = const()[name = string("input_343_groups_0"), val = int32(1024)]; - tensor input_343_strides_0 = const()[name = string("input_343_strides_0"), val = tensor([1, 1])]; - tensor input_343_dilations_0 = const()[name = string("input_343_dilations_0"), val = tensor([1, 1])]; - tensor const_292_to_fp16 = const()[name = string("const_292_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167150720)))]; - tensor const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167169216)))]; - tensor input_345_cast_fp16 = conv(bias = const_293_to_fp16, dilations = input_343_dilations_0, groups = input_343_groups_0, pad = input_343_pad_0, pad_type = input_343_pad_type_0, strides = input_343_strides_0, weight = const_292_to_fp16, x = input_341_cast_fp16)[name = string("input_345_cast_fp16")]; - tensor input_347_cast_fp16 = silu(x = input_345_cast_fp16)[name = string("input_347_cast_fp16")]; - string var_5003_pad_type_0 = const()[name = string("op_5003_pad_type_0"), val = string("valid")]; - tensor var_5003_strides_0 = const()[name = string("op_5003_strides_0"), val = tensor([1, 1])]; - tensor var_5003_pad_0 = const()[name = string("op_5003_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5003_dilations_0 = const()[name = string("op_5003_dilations_0"), val = tensor([1, 1])]; - int32 var_5003_groups_0 = const()[name = string("op_5003_groups_0"), val = int32(1)]; - tensor layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167171328))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167564608))))[name = string("layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5003_cast_fp16 = conv(dilations = var_5003_dilations_0, groups = var_5003_groups_0, pad = var_5003_pad_0, pad_type = var_5003_pad_type_0, strides = var_5003_strides_0, weight = layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_347_cast_fp16)[name = string("op_5003_cast_fp16")]; - string var_5009_pad_type_0 = const()[name = string("op_5009_pad_type_0"), val = string("valid")]; - tensor var_5009_strides_0 = const()[name = string("op_5009_strides_0"), val = tensor([1, 1])]; - tensor var_5009_pad_0 = const()[name = string("op_5009_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5009_dilations_0 = const()[name = string("op_5009_dilations_0"), val = tensor([1, 1])]; - int32 var_5009_groups_0 = const()[name = string("op_5009_groups_0"), val = int32(1)]; - tensor layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167575040))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167565696))))[name = string("layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5009_cast_fp16 = conv(dilations = var_5009_dilations_0, groups = var_5009_groups_0, pad = var_5009_pad_0, pad_type = var_5009_pad_type_0, strides = var_5009_strides_0, weight = layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_347_cast_fp16)[name = string("op_5009_cast_fp16")]; - tensor x_77_cast_fp16 = add(x = var_5003_cast_fp16, y = var_5009_cast_fp16)[name = string("x_77_cast_fp16")]; - tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = x_77_cast_fp16)[name = string("inputs_127_cast_fp16")]; - tensor out_127_axes_0 = const()[name = string("out_127_axes_0"), val = tensor([1])]; - fp16 var_5020_to_fp16 = const()[name = string("op_5020_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_5020_to_fp16, x = inputs_127_cast_fp16)[name = string("out_127_cast_fp16")]; - tensor input_349_gamma_0_to_fp16 = const()[name = string("input_349_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167706176)))]; - tensor input_349_beta_0_to_fp16 = const()[name = string("input_349_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167708288)))]; - fp16 input_349_epsilon_0_to_fp16 = const()[name = string("input_349_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_349_cast_fp16 = batch_norm(beta = input_349_beta_0_to_fp16, epsilon = input_349_epsilon_0_to_fp16, gamma = input_349_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_127_cast_fp16)[name = string("input_349_cast_fp16")]; - string var_5040_pad_type_0 = const()[name = string("op_5040_pad_type_0"), val = string("valid")]; - tensor var_5040_strides_0 = const()[name = string("op_5040_strides_0"), val = tensor([1, 1])]; - tensor var_5040_pad_0 = const()[name = string("op_5040_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5040_dilations_0 = const()[name = string("op_5040_dilations_0"), val = tensor([1, 1])]; - int32 var_5040_groups_0 = const()[name = string("op_5040_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167710400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169283328))))[name = string("layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5040_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5040_dilations_0, groups = var_5040_groups_0, pad = var_5040_pad_0, pad_type = var_5040_pad_type_0, strides = var_5040_strides_0, weight = layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_349_cast_fp16)[name = string("op_5040_cast_fp16")]; - string var_5046_pad_type_0 = const()[name = string("op_5046_pad_type_0"), val = string("valid")]; - tensor var_5046_strides_0 = const()[name = string("op_5046_strides_0"), val = tensor([1, 1])]; - tensor var_5046_pad_0 = const()[name = string("op_5046_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5046_dilations_0 = const()[name = string("op_5046_dilations_0"), val = tensor([1, 1])]; - int32 var_5046_groups_0 = const()[name = string("op_5046_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169332800))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169287488))))[name = string("layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5046_cast_fp16 = conv(dilations = var_5046_dilations_0, groups = var_5046_groups_0, pad = var_5046_pad_0, pad_type = var_5046_pad_type_0, strides = var_5046_strides_0, weight = layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_349_cast_fp16)[name = string("op_5046_cast_fp16")]; - tensor input_351_cast_fp16 = add(x = var_5040_cast_fp16, y = var_5046_cast_fp16)[name = string("input_351_cast_fp16")]; - tensor input_353_cast_fp16 = silu(x = input_351_cast_fp16)[name = string("input_353_cast_fp16")]; - string var_5057_pad_type_0 = const()[name = string("op_5057_pad_type_0"), val = string("valid")]; - tensor var_5057_strides_0 = const()[name = string("op_5057_strides_0"), val = tensor([1, 1])]; - tensor var_5057_pad_0 = const()[name = string("op_5057_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5057_dilations_0 = const()[name = string("op_5057_dilations_0"), val = tensor([1, 1])]; - int32 var_5057_groups_0 = const()[name = string("op_5057_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169857152))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171430080))))[name = string("layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5057_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5057_dilations_0, groups = var_5057_groups_0, pad = var_5057_pad_0, pad_type = var_5057_pad_type_0, strides = var_5057_strides_0, weight = layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_353_cast_fp16)[name = string("op_5057_cast_fp16")]; - string var_5063_pad_type_0 = const()[name = string("op_5063_pad_type_0"), val = string("valid")]; - tensor var_5063_strides_0 = const()[name = string("op_5063_strides_0"), val = tensor([1, 1])]; - tensor var_5063_pad_0 = const()[name = string("op_5063_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5063_dilations_0 = const()[name = string("op_5063_dilations_0"), val = tensor([1, 1])]; - int32 var_5063_groups_0 = const()[name = string("op_5063_groups_0"), val = int32(1)]; - tensor layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171487808))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171431168))))[name = string("layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5063_cast_fp16 = conv(dilations = var_5063_dilations_0, groups = var_5063_groups_0, pad = var_5063_pad_0, pad_type = var_5063_pad_type_0, strides = var_5063_strides_0, weight = layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_353_cast_fp16)[name = string("op_5063_cast_fp16")]; - tensor x_79_cast_fp16 = add(x = var_5057_cast_fp16, y = var_5063_cast_fp16)[name = string("x_79_cast_fp16")]; - fp16 var_5065_to_fp16 = const()[name = string("op_5065_to_fp16"), val = fp16(0x1p-1)]; - tensor var_5066_cast_fp16 = mul(x = x_79_cast_fp16, y = var_5065_to_fp16)[name = string("op_5066_cast_fp16")]; - tensor inputs_129_cast_fp16 = add(x = inputs_127_cast_fp16, y = var_5066_cast_fp16)[name = string("inputs_129_cast_fp16")]; - tensor out_129_axes_0 = const()[name = string("out_129_axes_0"), val = tensor([1])]; - fp16 var_5076_to_fp16 = const()[name = string("op_5076_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_129_cast_fp16 = layer_norm(axes = out_129_axes_0, epsilon = var_5076_to_fp16, x = inputs_129_cast_fp16)[name = string("out_129_cast_fp16")]; - tensor inputs_131_gamma_0_to_fp16 = const()[name = string("inputs_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172012160)))]; - tensor inputs_131_beta_0_to_fp16 = const()[name = string("inputs_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172014272)))]; - fp16 inputs_131_epsilon_0_to_fp16 = const()[name = string("inputs_131_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_131_cast_fp16 = batch_norm(beta = inputs_131_beta_0_to_fp16, epsilon = inputs_131_epsilon_0_to_fp16, gamma = inputs_131_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_129_cast_fp16)[name = string("inputs_131_cast_fp16")]; - int32 var_5090 = const()[name = string("op_5090"), val = int32(3)]; - tensor out_131_axes_0 = const()[name = string("out_131_axes_0"), val = tensor([1])]; - fp16 var_5121_to_fp16 = const()[name = string("op_5121_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_131_cast_fp16 = layer_norm(axes = out_131_axes_0, epsilon = var_5121_to_fp16, x = inputs_131_cast_fp16)[name = string("out_131_cast_fp16")]; - tensor input_355_gamma_0_to_fp16 = const()[name = string("input_355_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172016384)))]; - tensor input_355_beta_0_to_fp16 = const()[name = string("input_355_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172018496)))]; - fp16 input_355_epsilon_0_to_fp16 = const()[name = string("input_355_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_355_cast_fp16 = batch_norm(beta = input_355_beta_0_to_fp16, epsilon = input_355_epsilon_0_to_fp16, gamma = input_355_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_131_cast_fp16)[name = string("input_355_cast_fp16")]; - string var_5141_pad_type_0 = const()[name = string("op_5141_pad_type_0"), val = string("valid")]; - tensor var_5141_strides_0 = const()[name = string("op_5141_strides_0"), val = tensor([1, 1])]; - tensor var_5141_pad_0 = const()[name = string("op_5141_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5141_dilations_0 = const()[name = string("op_5141_dilations_0"), val = tensor([1, 1])]; - int32 var_5141_groups_0 = const()[name = string("op_5141_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172020608))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173593536))))[name = string("layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5141_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5141_dilations_0, groups = var_5141_groups_0, pad = var_5141_pad_0, pad_type = var_5141_pad_type_0, strides = var_5141_strides_0, weight = layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_355_cast_fp16)[name = string("op_5141_cast_fp16")]; - string var_5147_pad_type_0 = const()[name = string("op_5147_pad_type_0"), val = string("valid")]; - tensor var_5147_strides_0 = const()[name = string("op_5147_strides_0"), val = tensor([1, 1])]; - tensor var_5147_pad_0 = const()[name = string("op_5147_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5147_dilations_0 = const()[name = string("op_5147_dilations_0"), val = tensor([1, 1])]; - int32 var_5147_groups_0 = const()[name = string("op_5147_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173649600))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173597696))))[name = string("layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5147_cast_fp16 = conv(dilations = var_5147_dilations_0, groups = var_5147_groups_0, pad = var_5147_pad_0, pad_type = var_5147_pad_type_0, strides = var_5147_strides_0, weight = layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_355_cast_fp16)[name = string("op_5147_cast_fp16")]; - tensor input_357_cast_fp16 = add(x = var_5141_cast_fp16, y = var_5147_cast_fp16)[name = string("input_357_cast_fp16")]; - tensor input_359_cast_fp16 = silu(x = input_357_cast_fp16)[name = string("input_359_cast_fp16")]; - string var_5158_pad_type_0 = const()[name = string("op_5158_pad_type_0"), val = string("valid")]; - tensor var_5158_strides_0 = const()[name = string("op_5158_strides_0"), val = tensor([1, 1])]; - tensor var_5158_pad_0 = const()[name = string("op_5158_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5158_dilations_0 = const()[name = string("op_5158_dilations_0"), val = tensor([1, 1])]; - int32 var_5158_groups_0 = const()[name = string("op_5158_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174173952))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175746880))))[name = string("layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5158_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5158_dilations_0, groups = var_5158_groups_0, pad = var_5158_pad_0, pad_type = var_5158_pad_type_0, strides = var_5158_strides_0, weight = layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_359_cast_fp16)[name = string("op_5158_cast_fp16")]; - string var_5164_pad_type_0 = const()[name = string("op_5164_pad_type_0"), val = string("valid")]; - tensor var_5164_strides_0 = const()[name = string("op_5164_strides_0"), val = tensor([1, 1])]; - tensor var_5164_pad_0 = const()[name = string("op_5164_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5164_dilations_0 = const()[name = string("op_5164_dilations_0"), val = tensor([1, 1])]; - int32 var_5164_groups_0 = const()[name = string("op_5164_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175807232))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175747968))))[name = string("layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5164_cast_fp16 = conv(dilations = var_5164_dilations_0, groups = var_5164_groups_0, pad = var_5164_pad_0, pad_type = var_5164_pad_type_0, strides = var_5164_strides_0, weight = layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_359_cast_fp16)[name = string("op_5164_cast_fp16")]; - tensor x_81_cast_fp16 = add(x = var_5158_cast_fp16, y = var_5164_cast_fp16)[name = string("x_81_cast_fp16")]; - fp16 var_5166_to_fp16 = const()[name = string("op_5166_to_fp16"), val = fp16(0x1p-1)]; - tensor var_5167_cast_fp16 = mul(x = x_81_cast_fp16, y = var_5166_to_fp16)[name = string("op_5167_cast_fp16")]; - tensor inputs_133_cast_fp16 = add(x = inputs_131_cast_fp16, y = var_5167_cast_fp16)[name = string("inputs_133_cast_fp16")]; - tensor out_133_axes_0 = const()[name = string("out_133_axes_0"), val = tensor([1])]; - fp16 var_5177_to_fp16 = const()[name = string("op_5177_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_133_cast_fp16 = layer_norm(axes = out_133_axes_0, epsilon = var_5177_to_fp16, x = inputs_133_cast_fp16)[name = string("out_133_cast_fp16")]; - tensor obj_55_gamma_0_to_fp16 = const()[name = string("obj_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176331584)))]; - tensor obj_55_beta_0_to_fp16 = const()[name = string("obj_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176333696)))]; - fp16 obj_55_epsilon_0_to_fp16 = const()[name = string("obj_55_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_55_cast_fp16 = batch_norm(beta = obj_55_beta_0_to_fp16, epsilon = obj_55_epsilon_0_to_fp16, gamma = obj_55_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_133_cast_fp16)[name = string("obj_55_cast_fp16")]; - string var_5202_pad_type_0 = const()[name = string("op_5202_pad_type_0"), val = string("valid")]; - tensor var_5202_strides_0 = const()[name = string("op_5202_strides_0"), val = tensor([1, 1])]; - tensor var_5202_pad_0 = const()[name = string("op_5202_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5202_dilations_0 = const()[name = string("op_5202_dilations_0"), val = tensor([1, 1])]; - int32 var_5202_groups_0 = const()[name = string("op_5202_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176335808))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176729088))))[name = string("layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5202_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5202_dilations_0, groups = var_5202_groups_0, pad = var_5202_pad_0, pad_type = var_5202_pad_type_0, strides = var_5202_strides_0, weight = layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = string("op_5202_cast_fp16")]; - string var_5208_pad_type_0 = const()[name = string("op_5208_pad_type_0"), val = string("valid")]; - tensor var_5208_strides_0 = const()[name = string("op_5208_strides_0"), val = tensor([1, 1])]; - tensor var_5208_pad_0 = const()[name = string("op_5208_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5208_dilations_0 = const()[name = string("op_5208_dilations_0"), val = tensor([1, 1])]; - int32 var_5208_groups_0 = const()[name = string("op_5208_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176745536))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176730176))))[name = string("layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5208_cast_fp16 = conv(dilations = var_5208_dilations_0, groups = var_5208_groups_0, pad = var_5208_pad_0, pad_type = var_5208_pad_type_0, strides = var_5208_strides_0, weight = layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = string("op_5208_cast_fp16")]; - tensor query_53_cast_fp16 = add(x = var_5202_cast_fp16, y = var_5208_cast_fp16)[name = string("query_53_cast_fp16")]; - string var_5217_pad_type_0 = const()[name = string("op_5217_pad_type_0"), val = string("valid")]; - tensor var_5217_strides_0 = const()[name = string("op_5217_strides_0"), val = tensor([1, 1])]; - tensor var_5217_pad_0 = const()[name = string("op_5217_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5217_dilations_0 = const()[name = string("op_5217_dilations_0"), val = tensor([1, 1])]; - int32 var_5217_groups_0 = const()[name = string("op_5217_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176876672))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177269952))))[name = string("layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5217_cast_fp16 = conv(dilations = var_5217_dilations_0, groups = var_5217_groups_0, pad = var_5217_pad_0, pad_type = var_5217_pad_type_0, strides = var_5217_strides_0, weight = layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = string("op_5217_cast_fp16")]; - string var_5223_pad_type_0 = const()[name = string("op_5223_pad_type_0"), val = string("valid")]; - tensor var_5223_strides_0 = const()[name = string("op_5223_strides_0"), val = tensor([1, 1])]; - tensor var_5223_pad_0 = const()[name = string("op_5223_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5223_dilations_0 = const()[name = string("op_5223_dilations_0"), val = tensor([1, 1])]; - int32 var_5223_groups_0 = const()[name = string("op_5223_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177285440))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177271040))))[name = string("layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5223_cast_fp16 = conv(dilations = var_5223_dilations_0, groups = var_5223_groups_0, pad = var_5223_pad_0, pad_type = var_5223_pad_type_0, strides = var_5223_strides_0, weight = layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = string("op_5223_cast_fp16")]; - tensor key_27_cast_fp16 = add(x = var_5217_cast_fp16, y = var_5223_cast_fp16)[name = string("key_27_cast_fp16")]; - string var_5233_pad_type_0 = const()[name = string("op_5233_pad_type_0"), val = string("valid")]; - tensor var_5233_strides_0 = const()[name = string("op_5233_strides_0"), val = tensor([1, 1])]; - tensor var_5233_pad_0 = const()[name = string("op_5233_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5233_dilations_0 = const()[name = string("op_5233_dilations_0"), val = tensor([1, 1])]; - int32 var_5233_groups_0 = const()[name = string("op_5233_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177416576))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177809856))))[name = string("layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5233_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5233_dilations_0, groups = var_5233_groups_0, pad = var_5233_pad_0, pad_type = var_5233_pad_type_0, strides = var_5233_strides_0, weight = layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = string("op_5233_cast_fp16")]; - string var_5239_pad_type_0 = const()[name = string("op_5239_pad_type_0"), val = string("valid")]; - tensor var_5239_strides_0 = const()[name = string("op_5239_strides_0"), val = tensor([1, 1])]; - tensor var_5239_pad_0 = const()[name = string("op_5239_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5239_dilations_0 = const()[name = string("op_5239_dilations_0"), val = tensor([1, 1])]; - int32 var_5239_groups_0 = const()[name = string("op_5239_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177819328))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177810944))))[name = string("layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5239_cast_fp16 = conv(dilations = var_5239_dilations_0, groups = var_5239_groups_0, pad = var_5239_pad_0, pad_type = var_5239_pad_type_0, strides = var_5239_strides_0, weight = layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = string("op_5239_cast_fp16")]; - tensor value_27_cast_fp16 = add(x = var_5233_cast_fp16, y = var_5239_cast_fp16)[name = string("value_27_cast_fp16")]; - tensor var_5242_to_fp16 = const()[name = string("op_5242_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177950464)))]; - tensor query_55_cast_fp16 = add(x = query_53_cast_fp16, y = var_5242_to_fp16)[name = string("query_55_cast_fp16")]; - tensor var_5245_to_fp16 = const()[name = string("op_5245_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177952576)))]; - tensor q_with_bias_v_27_cast_fp16 = add(x = query_53_cast_fp16, y = var_5245_to_fp16)[name = string("q_with_bias_v_27_cast_fp16")]; - string var_5255_pad_type_0 = const()[name = string("op_5255_pad_type_0"), val = string("valid")]; - tensor var_5255_strides_0 = const()[name = string("op_5255_strides_0"), val = tensor([1, 1])]; - tensor var_5255_pad_0 = const()[name = string("op_5255_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5255_dilations_0 = const()[name = string("op_5255_dilations_0"), val = tensor([1, 1])]; - int32 var_5255_groups_0 = const()[name = string("op_5255_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177954688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178347968))))[name = string("layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_5255_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5255_dilations_0, groups = var_5255_groups_0, pad = var_5255_pad_0, pad_type = var_5255_pad_type_0, strides = var_5255_strides_0, weight = layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_5255_cast_fp16")]; - string var_5261_pad_type_0 = const()[name = string("op_5261_pad_type_0"), val = string("valid")]; - tensor var_5261_strides_0 = const()[name = string("op_5261_strides_0"), val = tensor([1, 1])]; - tensor var_5261_pad_0 = const()[name = string("op_5261_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5261_dilations_0 = const()[name = string("op_5261_dilations_0"), val = tensor([1, 1])]; - int32 var_5261_groups_0 = const()[name = string("op_5261_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178390208))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178349056))))[name = string("layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5261_cast_fp16 = conv(dilations = var_5261_dilations_0, groups = var_5261_groups_0, pad = var_5261_pad_0, pad_type = var_5261_pad_type_0, strides = var_5261_strides_0, weight = layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_5261_cast_fp16")]; - tensor p_27_cast_fp16 = add(x = var_5255_cast_fp16, y = var_5261_cast_fp16)[name = string("p_27_cast_fp16")]; - tensor var_5265 = const()[name = string("op_5265"), val = tensor([1, 8, 128, 188])]; - tensor var_5266_cast_fp16 = reshape(shape = var_5265, x = q_with_bias_v_27_cast_fp16)[name = string("op_5266_cast_fp16")]; - tensor var_5267 = const()[name = string("op_5267"), val = tensor([1, 8, 128, -1])]; - tensor var_5268_cast_fp16 = reshape(shape = var_5267, x = p_27_cast_fp16)[name = string("op_5268_cast_fp16")]; - bool matrix_bd_105_transpose_x_0 = const()[name = string("matrix_bd_105_transpose_x_0"), val = bool(true)]; - bool matrix_bd_105_transpose_y_0 = const()[name = string("matrix_bd_105_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_105_cast_fp16 = matmul(transpose_x = matrix_bd_105_transpose_x_0, transpose_y = matrix_bd_105_transpose_y_0, x = var_5266_cast_fp16, y = var_5268_cast_fp16)[name = string("matrix_bd_105_cast_fp16")]; - tensor matrix_bd_107_pad_0 = const()[name = string("matrix_bd_107_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_107_mode_0 = const()[name = string("matrix_bd_107_mode_0"), val = string("constant")]; - fp16 const_153_to_fp16 = const()[name = string("const_153_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_107_cast_fp16 = pad(constant_val = const_153_to_fp16, mode = matrix_bd_107_mode_0, pad = matrix_bd_107_pad_0, x = matrix_bd_105_cast_fp16)[name = string("matrix_bd_107_cast_fp16")]; - tensor var_5277 = const()[name = string("op_5277"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_109_cast_fp16 = reshape(shape = var_5277, x = matrix_bd_107_cast_fp16)[name = string("matrix_bd_109_cast_fp16")]; - tensor var_5281_begin_0 = const()[name = string("op_5281_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_5281_end_0 = const()[name = string("op_5281_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_5281_end_mask_0 = const()[name = string("op_5281_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_5281_cast_fp16 = slice_by_index(begin = var_5281_begin_0, end = var_5281_end_0, end_mask = var_5281_end_mask_0, x = matrix_bd_109_cast_fp16)[name = string("op_5281_cast_fp16")]; - tensor var_5282 = const()[name = string("op_5282"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_111_cast_fp16 = reshape(shape = var_5282, x = var_5281_cast_fp16)[name = string("matrix_bd_111_cast_fp16")]; - tensor var_5287_begin_0 = const()[name = string("op_5287_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5287_end_0 = const()[name = string("op_5287_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_5287_end_mask_0 = const()[name = string("op_5287_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_5287_cast_fp16 = slice_by_index(begin = var_5287_begin_0, end = var_5287_end_0, end_mask = var_5287_end_mask_0, x = matrix_bd_111_cast_fp16)[name = string("op_5287_cast_fp16")]; - fp16 var_5288_to_fp16 = const()[name = string("op_5288_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_27_cast_fp16 = mul(x = var_5287_cast_fp16, y = var_5288_to_fp16)[name = string("qk_mask_27_cast_fp16")]; - tensor var_5292 = const()[name = string("op_5292"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_27_cast_fp16 = reshape(shape = var_5292, x = query_55_cast_fp16)[name = string("mh_q_27_cast_fp16")]; - fp16 var_5294_to_fp16 = const()[name = string("op_5294_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_5295_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_5294_to_fp16)[name = string("op_5295_cast_fp16")]; - tensor var_5298 = const()[name = string("op_5298"), val = tensor([1, 8, 128, 188])]; - tensor var_5299_cast_fp16 = reshape(shape = var_5298, x = key_27_cast_fp16)[name = string("op_5299_cast_fp16")]; - bool mh_w_53_transpose_x_0 = const()[name = string("mh_w_53_transpose_x_0"), val = bool(true)]; - bool mh_w_53_transpose_y_0 = const()[name = string("mh_w_53_transpose_y_0"), val = bool(false)]; - tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_5295_cast_fp16, y = var_5299_cast_fp16)[name = string("mh_w_53_cast_fp16")]; - tensor mh_w_55_cast_fp16 = add(x = mh_w_53_cast_fp16, y = qk_mask_27_cast_fp16)[name = string("mh_w_55_cast_fp16")]; - tensor var_5303_cast_fp16 = softmax(axis = var_5090, x = mh_w_55_cast_fp16)[name = string("op_5303_cast_fp16")]; - tensor var_5304 = const()[name = string("op_5304"), val = tensor([1, 8, 128, 188])]; - tensor var_5305_cast_fp16 = reshape(shape = var_5304, x = value_27_cast_fp16)[name = string("op_5305_cast_fp16")]; - bool attn_27_transpose_x_0 = const()[name = string("attn_27_transpose_x_0"), val = bool(false)]; - bool attn_27_transpose_y_0 = const()[name = string("attn_27_transpose_y_0"), val = bool(true)]; - tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_5305_cast_fp16, y = var_5303_cast_fp16)[name = string("attn_27_cast_fp16")]; - tensor var_5308 = const()[name = string("op_5308"), val = tensor([1, 1024, 1, 188])]; - tensor input_361_cast_fp16 = reshape(shape = var_5308, x = attn_27_cast_fp16)[name = string("input_361_cast_fp16")]; - string var_5318_pad_type_0 = const()[name = string("op_5318_pad_type_0"), val = string("valid")]; - tensor var_5318_strides_0 = const()[name = string("op_5318_strides_0"), val = tensor([1, 1])]; - tensor var_5318_pad_0 = const()[name = string("op_5318_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5318_dilations_0 = const()[name = string("op_5318_dilations_0"), val = tensor([1, 1])]; - int32 var_5318_groups_0 = const()[name = string("op_5318_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178521344))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178914624))))[name = string("layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5318_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5318_dilations_0, groups = var_5318_groups_0, pad = var_5318_pad_0, pad_type = var_5318_pad_type_0, strides = var_5318_strides_0, weight = layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_361_cast_fp16)[name = string("op_5318_cast_fp16")]; - string var_5324_pad_type_0 = const()[name = string("op_5324_pad_type_0"), val = string("valid")]; - tensor var_5324_strides_0 = const()[name = string("op_5324_strides_0"), val = tensor([1, 1])]; - tensor var_5324_pad_0 = const()[name = string("op_5324_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5324_dilations_0 = const()[name = string("op_5324_dilations_0"), val = tensor([1, 1])]; - int32 var_5324_groups_0 = const()[name = string("op_5324_groups_0"), val = int32(1)]; - tensor layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178924032))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178915712))))[name = string("layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5324_cast_fp16 = conv(dilations = var_5324_dilations_0, groups = var_5324_groups_0, pad = var_5324_pad_0, pad_type = var_5324_pad_type_0, strides = var_5324_strides_0, weight = layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_361_cast_fp16)[name = string("op_5324_cast_fp16")]; - tensor obj_57_cast_fp16 = add(x = var_5318_cast_fp16, y = var_5324_cast_fp16)[name = string("obj_57_cast_fp16")]; - tensor inputs_135_cast_fp16 = add(x = inputs_133_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_135_cast_fp16")]; - tensor out_135_axes_0 = const()[name = string("out_135_axes_0"), val = tensor([1])]; - fp16 var_5335_to_fp16 = const()[name = string("op_5335_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_135_cast_fp16 = layer_norm(axes = out_135_axes_0, epsilon = var_5335_to_fp16, x = inputs_135_cast_fp16)[name = string("out_135_cast_fp16")]; - tensor input_363_gamma_0_to_fp16 = const()[name = string("input_363_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179055168)))]; - tensor input_363_beta_0_to_fp16 = const()[name = string("input_363_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179057280)))]; - fp16 input_363_epsilon_0_to_fp16 = const()[name = string("input_363_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_363_cast_fp16 = batch_norm(beta = input_363_beta_0_to_fp16, epsilon = input_363_epsilon_0_to_fp16, gamma = input_363_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_135_cast_fp16)[name = string("input_363_cast_fp16")]; - string var_5356_pad_type_0 = const()[name = string("op_5356_pad_type_0"), val = string("valid")]; - tensor var_5356_strides_0 = const()[name = string("op_5356_strides_0"), val = tensor([1, 1])]; - tensor var_5356_pad_0 = const()[name = string("op_5356_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5356_dilations_0 = const()[name = string("op_5356_dilations_0"), val = tensor([1, 1])]; - int32 var_5356_groups_0 = const()[name = string("op_5356_groups_0"), val = int32(1)]; - tensor layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179059392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179845888))))[name = string("layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5356_cast_fp16 = conv(dilations = var_5356_dilations_0, groups = var_5356_groups_0, pad = var_5356_pad_0, pad_type = var_5356_pad_type_0, strides = var_5356_strides_0, weight = layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_363_cast_fp16)[name = string("op_5356_cast_fp16")]; - string var_5362_pad_type_0 = const()[name = string("op_5362_pad_type_0"), val = string("valid")]; - tensor var_5362_strides_0 = const()[name = string("op_5362_strides_0"), val = tensor([1, 1])]; - tensor var_5362_pad_0 = const()[name = string("op_5362_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5362_dilations_0 = const()[name = string("op_5362_dilations_0"), val = tensor([1, 1])]; - int32 var_5362_groups_0 = const()[name = string("op_5362_groups_0"), val = int32(1)]; - tensor layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179867840))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179848000))))[name = string("layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5362_cast_fp16 = conv(dilations = var_5362_dilations_0, groups = var_5362_groups_0, pad = var_5362_pad_0, pad_type = var_5362_pad_type_0, strides = var_5362_strides_0, weight = layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_363_cast_fp16)[name = string("op_5362_cast_fp16")]; - tensor input_365_cast_fp16 = add(x = var_5356_cast_fp16, y = var_5362_cast_fp16)[name = string("input_365_cast_fp16")]; - int32 input_367_split_num_splits_0 = const()[name = string("input_367_split_num_splits_0"), val = int32(2)]; - int32 input_367_split_axis_0 = const()[name = string("input_367_split_axis_0"), val = int32(1)]; - tensor input_367_split_cast_fp16_0, tensor input_367_split_cast_fp16_1 = split(axis = input_367_split_axis_0, num_splits = input_367_split_num_splits_0, x = input_365_cast_fp16)[name = string("input_367_split_cast_fp16")]; - tensor input_367_split_1_sigmoid_cast_fp16 = sigmoid(x = input_367_split_cast_fp16_1)[name = string("input_367_split_1_sigmoid_cast_fp16")]; - tensor input_367_cast_fp16 = mul(x = input_367_split_cast_fp16_0, y = input_367_split_1_sigmoid_cast_fp16)[name = string("input_367_cast_fp16")]; - string input_369_pad_type_0 = const()[name = string("input_369_pad_type_0"), val = string("custom")]; - tensor input_369_pad_0 = const()[name = string("input_369_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_369_groups_0 = const()[name = string("input_369_groups_0"), val = int32(1024)]; - tensor input_369_strides_0 = const()[name = string("input_369_strides_0"), val = tensor([1, 1])]; - tensor input_369_dilations_0 = const()[name = string("input_369_dilations_0"), val = tensor([1, 1])]; - tensor const_294_to_fp16 = const()[name = string("const_294_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180130048)))]; - tensor const_295_to_fp16 = const()[name = string("const_295_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180148544)))]; - tensor input_371_cast_fp16 = conv(bias = const_295_to_fp16, dilations = input_369_dilations_0, groups = input_369_groups_0, pad = input_369_pad_0, pad_type = input_369_pad_type_0, strides = input_369_strides_0, weight = const_294_to_fp16, x = input_367_cast_fp16)[name = string("input_371_cast_fp16")]; - tensor input_373_cast_fp16 = silu(x = input_371_cast_fp16)[name = string("input_373_cast_fp16")]; - string var_5384_pad_type_0 = const()[name = string("op_5384_pad_type_0"), val = string("valid")]; - tensor var_5384_strides_0 = const()[name = string("op_5384_strides_0"), val = tensor([1, 1])]; - tensor var_5384_pad_0 = const()[name = string("op_5384_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5384_dilations_0 = const()[name = string("op_5384_dilations_0"), val = tensor([1, 1])]; - int32 var_5384_groups_0 = const()[name = string("op_5384_groups_0"), val = int32(1)]; - tensor layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180150656))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180543936))))[name = string("layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5384_cast_fp16 = conv(dilations = var_5384_dilations_0, groups = var_5384_groups_0, pad = var_5384_pad_0, pad_type = var_5384_pad_type_0, strides = var_5384_strides_0, weight = layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_373_cast_fp16)[name = string("op_5384_cast_fp16")]; - string var_5390_pad_type_0 = const()[name = string("op_5390_pad_type_0"), val = string("valid")]; - tensor var_5390_strides_0 = const()[name = string("op_5390_strides_0"), val = tensor([1, 1])]; - tensor var_5390_pad_0 = const()[name = string("op_5390_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5390_dilations_0 = const()[name = string("op_5390_dilations_0"), val = tensor([1, 1])]; - int32 var_5390_groups_0 = const()[name = string("op_5390_groups_0"), val = int32(1)]; - tensor layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180553984))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180545024))))[name = string("layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5390_cast_fp16 = conv(dilations = var_5390_dilations_0, groups = var_5390_groups_0, pad = var_5390_pad_0, pad_type = var_5390_pad_type_0, strides = var_5390_strides_0, weight = layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_373_cast_fp16)[name = string("op_5390_cast_fp16")]; - tensor x_83_cast_fp16 = add(x = var_5384_cast_fp16, y = var_5390_cast_fp16)[name = string("x_83_cast_fp16")]; - tensor inputs_137_cast_fp16 = add(x = inputs_135_cast_fp16, y = x_83_cast_fp16)[name = string("inputs_137_cast_fp16")]; - tensor out_137_axes_0 = const()[name = string("out_137_axes_0"), val = tensor([1])]; - fp16 var_5401_to_fp16 = const()[name = string("op_5401_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_137_cast_fp16 = layer_norm(axes = out_137_axes_0, epsilon = var_5401_to_fp16, x = inputs_137_cast_fp16)[name = string("out_137_cast_fp16")]; - tensor input_375_gamma_0_to_fp16 = const()[name = string("input_375_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180685120)))]; - tensor input_375_beta_0_to_fp16 = const()[name = string("input_375_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180687232)))]; - fp16 input_375_epsilon_0_to_fp16 = const()[name = string("input_375_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_375_cast_fp16 = batch_norm(beta = input_375_beta_0_to_fp16, epsilon = input_375_epsilon_0_to_fp16, gamma = input_375_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_137_cast_fp16)[name = string("input_375_cast_fp16")]; - string var_5421_pad_type_0 = const()[name = string("op_5421_pad_type_0"), val = string("valid")]; - tensor var_5421_strides_0 = const()[name = string("op_5421_strides_0"), val = tensor([1, 1])]; - tensor var_5421_pad_0 = const()[name = string("op_5421_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5421_dilations_0 = const()[name = string("op_5421_dilations_0"), val = tensor([1, 1])]; - int32 var_5421_groups_0 = const()[name = string("op_5421_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180689344))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182262272))))[name = string("layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5421_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5421_dilations_0, groups = var_5421_groups_0, pad = var_5421_pad_0, pad_type = var_5421_pad_type_0, strides = var_5421_strides_0, weight = layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_375_cast_fp16)[name = string("op_5421_cast_fp16")]; - string var_5427_pad_type_0 = const()[name = string("op_5427_pad_type_0"), val = string("valid")]; - tensor var_5427_strides_0 = const()[name = string("op_5427_strides_0"), val = tensor([1, 1])]; - tensor var_5427_pad_0 = const()[name = string("op_5427_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5427_dilations_0 = const()[name = string("op_5427_dilations_0"), val = tensor([1, 1])]; - int32 var_5427_groups_0 = const()[name = string("op_5427_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182309568))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182266432))))[name = string("layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5427_cast_fp16 = conv(dilations = var_5427_dilations_0, groups = var_5427_groups_0, pad = var_5427_pad_0, pad_type = var_5427_pad_type_0, strides = var_5427_strides_0, weight = layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_375_cast_fp16)[name = string("op_5427_cast_fp16")]; - tensor input_377_cast_fp16 = add(x = var_5421_cast_fp16, y = var_5427_cast_fp16)[name = string("input_377_cast_fp16")]; - tensor input_379_cast_fp16 = silu(x = input_377_cast_fp16)[name = string("input_379_cast_fp16")]; - string var_5438_pad_type_0 = const()[name = string("op_5438_pad_type_0"), val = string("valid")]; - tensor var_5438_strides_0 = const()[name = string("op_5438_strides_0"), val = tensor([1, 1])]; - tensor var_5438_pad_0 = const()[name = string("op_5438_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5438_dilations_0 = const()[name = string("op_5438_dilations_0"), val = tensor([1, 1])]; - int32 var_5438_groups_0 = const()[name = string("op_5438_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182833920))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184406848))))[name = string("layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5438_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5438_dilations_0, groups = var_5438_groups_0, pad = var_5438_pad_0, pad_type = var_5438_pad_type_0, strides = var_5438_strides_0, weight = layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_379_cast_fp16)[name = string("op_5438_cast_fp16")]; - string var_5444_pad_type_0 = const()[name = string("op_5444_pad_type_0"), val = string("valid")]; - tensor var_5444_strides_0 = const()[name = string("op_5444_strides_0"), val = tensor([1, 1])]; - tensor var_5444_pad_0 = const()[name = string("op_5444_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5444_dilations_0 = const()[name = string("op_5444_dilations_0"), val = tensor([1, 1])]; - int32 var_5444_groups_0 = const()[name = string("op_5444_groups_0"), val = int32(1)]; - tensor layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184462656))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184407936))))[name = string("layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5444_cast_fp16 = conv(dilations = var_5444_dilations_0, groups = var_5444_groups_0, pad = var_5444_pad_0, pad_type = var_5444_pad_type_0, strides = var_5444_strides_0, weight = layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_379_cast_fp16)[name = string("op_5444_cast_fp16")]; - tensor x_85_cast_fp16 = add(x = var_5438_cast_fp16, y = var_5444_cast_fp16)[name = string("x_85_cast_fp16")]; - fp16 var_5446_to_fp16 = const()[name = string("op_5446_to_fp16"), val = fp16(0x1p-1)]; - tensor var_5447_cast_fp16 = mul(x = x_85_cast_fp16, y = var_5446_to_fp16)[name = string("op_5447_cast_fp16")]; - tensor inputs_139_cast_fp16 = add(x = inputs_137_cast_fp16, y = var_5447_cast_fp16)[name = string("inputs_139_cast_fp16")]; - tensor out_139_axes_0 = const()[name = string("out_139_axes_0"), val = tensor([1])]; - fp16 var_5457_to_fp16 = const()[name = string("op_5457_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_139_cast_fp16 = layer_norm(axes = out_139_axes_0, epsilon = var_5457_to_fp16, x = inputs_139_cast_fp16)[name = string("out_139_cast_fp16")]; - tensor inputs_141_gamma_0_to_fp16 = const()[name = string("inputs_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184987008)))]; - tensor inputs_141_beta_0_to_fp16 = const()[name = string("inputs_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184989120)))]; - fp16 inputs_141_epsilon_0_to_fp16 = const()[name = string("inputs_141_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_141_cast_fp16 = batch_norm(beta = inputs_141_beta_0_to_fp16, epsilon = inputs_141_epsilon_0_to_fp16, gamma = inputs_141_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_139_cast_fp16)[name = string("inputs_141_cast_fp16")]; - int32 var_5471 = const()[name = string("op_5471"), val = int32(3)]; - tensor out_141_axes_0 = const()[name = string("out_141_axes_0"), val = tensor([1])]; - fp16 var_5502_to_fp16 = const()[name = string("op_5502_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_141_cast_fp16 = layer_norm(axes = out_141_axes_0, epsilon = var_5502_to_fp16, x = inputs_141_cast_fp16)[name = string("out_141_cast_fp16")]; - tensor input_381_gamma_0_to_fp16 = const()[name = string("input_381_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184991232)))]; - tensor input_381_beta_0_to_fp16 = const()[name = string("input_381_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184993344)))]; - fp16 input_381_epsilon_0_to_fp16 = const()[name = string("input_381_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_381_cast_fp16 = batch_norm(beta = input_381_beta_0_to_fp16, epsilon = input_381_epsilon_0_to_fp16, gamma = input_381_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_141_cast_fp16)[name = string("input_381_cast_fp16")]; - string var_5522_pad_type_0 = const()[name = string("op_5522_pad_type_0"), val = string("valid")]; - tensor var_5522_strides_0 = const()[name = string("op_5522_strides_0"), val = tensor([1, 1])]; - tensor var_5522_pad_0 = const()[name = string("op_5522_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5522_dilations_0 = const()[name = string("op_5522_dilations_0"), val = tensor([1, 1])]; - int32 var_5522_groups_0 = const()[name = string("op_5522_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184995456))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186568384))))[name = string("layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5522_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5522_dilations_0, groups = var_5522_groups_0, pad = var_5522_pad_0, pad_type = var_5522_pad_type_0, strides = var_5522_strides_0, weight = layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_381_cast_fp16)[name = string("op_5522_cast_fp16")]; - string var_5528_pad_type_0 = const()[name = string("op_5528_pad_type_0"), val = string("valid")]; - tensor var_5528_strides_0 = const()[name = string("op_5528_strides_0"), val = tensor([1, 1])]; - tensor var_5528_pad_0 = const()[name = string("op_5528_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5528_dilations_0 = const()[name = string("op_5528_dilations_0"), val = tensor([1, 1])]; - int32 var_5528_groups_0 = const()[name = string("op_5528_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186622208))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186572544))))[name = string("layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5528_cast_fp16 = conv(dilations = var_5528_dilations_0, groups = var_5528_groups_0, pad = var_5528_pad_0, pad_type = var_5528_pad_type_0, strides = var_5528_strides_0, weight = layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_381_cast_fp16)[name = string("op_5528_cast_fp16")]; - tensor input_383_cast_fp16 = add(x = var_5522_cast_fp16, y = var_5528_cast_fp16)[name = string("input_383_cast_fp16")]; - tensor input_385_cast_fp16 = silu(x = input_383_cast_fp16)[name = string("input_385_cast_fp16")]; - string var_5539_pad_type_0 = const()[name = string("op_5539_pad_type_0"), val = string("valid")]; - tensor var_5539_strides_0 = const()[name = string("op_5539_strides_0"), val = tensor([1, 1])]; - tensor var_5539_pad_0 = const()[name = string("op_5539_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5539_dilations_0 = const()[name = string("op_5539_dilations_0"), val = tensor([1, 1])]; - int32 var_5539_groups_0 = const()[name = string("op_5539_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187146560))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188719488))))[name = string("layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5539_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5539_dilations_0, groups = var_5539_groups_0, pad = var_5539_pad_0, pad_type = var_5539_pad_type_0, strides = var_5539_strides_0, weight = layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_385_cast_fp16)[name = string("op_5539_cast_fp16")]; - string var_5545_pad_type_0 = const()[name = string("op_5545_pad_type_0"), val = string("valid")]; - tensor var_5545_strides_0 = const()[name = string("op_5545_strides_0"), val = tensor([1, 1])]; - tensor var_5545_pad_0 = const()[name = string("op_5545_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5545_dilations_0 = const()[name = string("op_5545_dilations_0"), val = tensor([1, 1])]; - int32 var_5545_groups_0 = const()[name = string("op_5545_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188781888))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188720576))))[name = string("layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5545_cast_fp16 = conv(dilations = var_5545_dilations_0, groups = var_5545_groups_0, pad = var_5545_pad_0, pad_type = var_5545_pad_type_0, strides = var_5545_strides_0, weight = layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_385_cast_fp16)[name = string("op_5545_cast_fp16")]; - tensor x_87_cast_fp16 = add(x = var_5539_cast_fp16, y = var_5545_cast_fp16)[name = string("x_87_cast_fp16")]; - fp16 var_5547_to_fp16 = const()[name = string("op_5547_to_fp16"), val = fp16(0x1p-1)]; - tensor var_5548_cast_fp16 = mul(x = x_87_cast_fp16, y = var_5547_to_fp16)[name = string("op_5548_cast_fp16")]; - tensor inputs_143_cast_fp16 = add(x = inputs_141_cast_fp16, y = var_5548_cast_fp16)[name = string("inputs_143_cast_fp16")]; - tensor out_143_axes_0 = const()[name = string("out_143_axes_0"), val = tensor([1])]; - fp16 var_5558_to_fp16 = const()[name = string("op_5558_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_143_cast_fp16 = layer_norm(axes = out_143_axes_0, epsilon = var_5558_to_fp16, x = inputs_143_cast_fp16)[name = string("out_143_cast_fp16")]; - tensor obj_59_gamma_0_to_fp16 = const()[name = string("obj_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189306240)))]; - tensor obj_59_beta_0_to_fp16 = const()[name = string("obj_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189308352)))]; - fp16 obj_59_epsilon_0_to_fp16 = const()[name = string("obj_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_59_cast_fp16 = batch_norm(beta = obj_59_beta_0_to_fp16, epsilon = obj_59_epsilon_0_to_fp16, gamma = obj_59_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_143_cast_fp16)[name = string("obj_59_cast_fp16")]; - string var_5583_pad_type_0 = const()[name = string("op_5583_pad_type_0"), val = string("valid")]; - tensor var_5583_strides_0 = const()[name = string("op_5583_strides_0"), val = tensor([1, 1])]; - tensor var_5583_pad_0 = const()[name = string("op_5583_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5583_dilations_0 = const()[name = string("op_5583_dilations_0"), val = tensor([1, 1])]; - int32 var_5583_groups_0 = const()[name = string("op_5583_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189310464))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189703744))))[name = string("layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5583_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5583_dilations_0, groups = var_5583_groups_0, pad = var_5583_pad_0, pad_type = var_5583_pad_type_0, strides = var_5583_strides_0, weight = layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = string("op_5583_cast_fp16")]; - string var_5589_pad_type_0 = const()[name = string("op_5589_pad_type_0"), val = string("valid")]; - tensor var_5589_strides_0 = const()[name = string("op_5589_strides_0"), val = tensor([1, 1])]; - tensor var_5589_pad_0 = const()[name = string("op_5589_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5589_dilations_0 = const()[name = string("op_5589_dilations_0"), val = tensor([1, 1])]; - int32 var_5589_groups_0 = const()[name = string("op_5589_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189718592))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189704832))))[name = string("layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5589_cast_fp16 = conv(dilations = var_5589_dilations_0, groups = var_5589_groups_0, pad = var_5589_pad_0, pad_type = var_5589_pad_type_0, strides = var_5589_strides_0, weight = layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = string("op_5589_cast_fp16")]; - tensor query_57_cast_fp16 = add(x = var_5583_cast_fp16, y = var_5589_cast_fp16)[name = string("query_57_cast_fp16")]; - string var_5598_pad_type_0 = const()[name = string("op_5598_pad_type_0"), val = string("valid")]; - tensor var_5598_strides_0 = const()[name = string("op_5598_strides_0"), val = tensor([1, 1])]; - tensor var_5598_pad_0 = const()[name = string("op_5598_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5598_dilations_0 = const()[name = string("op_5598_dilations_0"), val = tensor([1, 1])]; - int32 var_5598_groups_0 = const()[name = string("op_5598_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189849728))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190243008))))[name = string("layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5598_cast_fp16 = conv(dilations = var_5598_dilations_0, groups = var_5598_groups_0, pad = var_5598_pad_0, pad_type = var_5598_pad_type_0, strides = var_5598_strides_0, weight = layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = string("op_5598_cast_fp16")]; - string var_5604_pad_type_0 = const()[name = string("op_5604_pad_type_0"), val = string("valid")]; - tensor var_5604_strides_0 = const()[name = string("op_5604_strides_0"), val = tensor([1, 1])]; - tensor var_5604_pad_0 = const()[name = string("op_5604_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5604_dilations_0 = const()[name = string("op_5604_dilations_0"), val = tensor([1, 1])]; - int32 var_5604_groups_0 = const()[name = string("op_5604_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190257664))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190244096))))[name = string("layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5604_cast_fp16 = conv(dilations = var_5604_dilations_0, groups = var_5604_groups_0, pad = var_5604_pad_0, pad_type = var_5604_pad_type_0, strides = var_5604_strides_0, weight = layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = string("op_5604_cast_fp16")]; - tensor key_29_cast_fp16 = add(x = var_5598_cast_fp16, y = var_5604_cast_fp16)[name = string("key_29_cast_fp16")]; - string var_5614_pad_type_0 = const()[name = string("op_5614_pad_type_0"), val = string("valid")]; - tensor var_5614_strides_0 = const()[name = string("op_5614_strides_0"), val = tensor([1, 1])]; - tensor var_5614_pad_0 = const()[name = string("op_5614_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5614_dilations_0 = const()[name = string("op_5614_dilations_0"), val = tensor([1, 1])]; - int32 var_5614_groups_0 = const()[name = string("op_5614_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190388800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190782080))))[name = string("layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5614_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5614_dilations_0, groups = var_5614_groups_0, pad = var_5614_pad_0, pad_type = var_5614_pad_type_0, strides = var_5614_strides_0, weight = layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = string("op_5614_cast_fp16")]; - string var_5620_pad_type_0 = const()[name = string("op_5620_pad_type_0"), val = string("valid")]; - tensor var_5620_strides_0 = const()[name = string("op_5620_strides_0"), val = tensor([1, 1])]; - tensor var_5620_pad_0 = const()[name = string("op_5620_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5620_dilations_0 = const()[name = string("op_5620_dilations_0"), val = tensor([1, 1])]; - int32 var_5620_groups_0 = const()[name = string("op_5620_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190792192))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190783168))))[name = string("layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5620_cast_fp16 = conv(dilations = var_5620_dilations_0, groups = var_5620_groups_0, pad = var_5620_pad_0, pad_type = var_5620_pad_type_0, strides = var_5620_strides_0, weight = layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = string("op_5620_cast_fp16")]; - tensor value_29_cast_fp16 = add(x = var_5614_cast_fp16, y = var_5620_cast_fp16)[name = string("value_29_cast_fp16")]; - tensor var_5623_to_fp16 = const()[name = string("op_5623_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190923328)))]; - tensor query_59_cast_fp16 = add(x = query_57_cast_fp16, y = var_5623_to_fp16)[name = string("query_59_cast_fp16")]; - tensor var_5626_to_fp16 = const()[name = string("op_5626_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190925440)))]; - tensor q_with_bias_v_29_cast_fp16 = add(x = query_57_cast_fp16, y = var_5626_to_fp16)[name = string("q_with_bias_v_29_cast_fp16")]; - string var_5636_pad_type_0 = const()[name = string("op_5636_pad_type_0"), val = string("valid")]; - tensor var_5636_strides_0 = const()[name = string("op_5636_strides_0"), val = tensor([1, 1])]; - tensor var_5636_pad_0 = const()[name = string("op_5636_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5636_dilations_0 = const()[name = string("op_5636_dilations_0"), val = tensor([1, 1])]; - int32 var_5636_groups_0 = const()[name = string("op_5636_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190927552))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191320832))))[name = string("layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_5636_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5636_dilations_0, groups = var_5636_groups_0, pad = var_5636_pad_0, pad_type = var_5636_pad_type_0, strides = var_5636_strides_0, weight = layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_5636_cast_fp16")]; - string var_5642_pad_type_0 = const()[name = string("op_5642_pad_type_0"), val = string("valid")]; - tensor var_5642_strides_0 = const()[name = string("op_5642_strides_0"), val = tensor([1, 1])]; - tensor var_5642_pad_0 = const()[name = string("op_5642_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5642_dilations_0 = const()[name = string("op_5642_dilations_0"), val = tensor([1, 1])]; - int32 var_5642_groups_0 = const()[name = string("op_5642_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191363008))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191321920))))[name = string("layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5642_cast_fp16 = conv(dilations = var_5642_dilations_0, groups = var_5642_groups_0, pad = var_5642_pad_0, pad_type = var_5642_pad_type_0, strides = var_5642_strides_0, weight = layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_5642_cast_fp16")]; - tensor p_29_cast_fp16 = add(x = var_5636_cast_fp16, y = var_5642_cast_fp16)[name = string("p_29_cast_fp16")]; - tensor var_5646 = const()[name = string("op_5646"), val = tensor([1, 8, 128, 188])]; - tensor var_5647_cast_fp16 = reshape(shape = var_5646, x = q_with_bias_v_29_cast_fp16)[name = string("op_5647_cast_fp16")]; - tensor var_5648 = const()[name = string("op_5648"), val = tensor([1, 8, 128, -1])]; - tensor var_5649_cast_fp16 = reshape(shape = var_5648, x = p_29_cast_fp16)[name = string("op_5649_cast_fp16")]; - bool matrix_bd_113_transpose_x_0 = const()[name = string("matrix_bd_113_transpose_x_0"), val = bool(true)]; - bool matrix_bd_113_transpose_y_0 = const()[name = string("matrix_bd_113_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_113_cast_fp16 = matmul(transpose_x = matrix_bd_113_transpose_x_0, transpose_y = matrix_bd_113_transpose_y_0, x = var_5647_cast_fp16, y = var_5649_cast_fp16)[name = string("matrix_bd_113_cast_fp16")]; - tensor matrix_bd_115_pad_0 = const()[name = string("matrix_bd_115_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_115_mode_0 = const()[name = string("matrix_bd_115_mode_0"), val = string("constant")]; - fp16 const_164_to_fp16 = const()[name = string("const_164_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_115_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = matrix_bd_115_mode_0, pad = matrix_bd_115_pad_0, x = matrix_bd_113_cast_fp16)[name = string("matrix_bd_115_cast_fp16")]; - tensor var_5658 = const()[name = string("op_5658"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_117_cast_fp16 = reshape(shape = var_5658, x = matrix_bd_115_cast_fp16)[name = string("matrix_bd_117_cast_fp16")]; - tensor var_5662_begin_0 = const()[name = string("op_5662_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_5662_end_0 = const()[name = string("op_5662_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_5662_end_mask_0 = const()[name = string("op_5662_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_5662_cast_fp16 = slice_by_index(begin = var_5662_begin_0, end = var_5662_end_0, end_mask = var_5662_end_mask_0, x = matrix_bd_117_cast_fp16)[name = string("op_5662_cast_fp16")]; - tensor var_5663 = const()[name = string("op_5663"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_119_cast_fp16 = reshape(shape = var_5663, x = var_5662_cast_fp16)[name = string("matrix_bd_119_cast_fp16")]; - tensor var_5668_begin_0 = const()[name = string("op_5668_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5668_end_0 = const()[name = string("op_5668_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_5668_end_mask_0 = const()[name = string("op_5668_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_5668_cast_fp16 = slice_by_index(begin = var_5668_begin_0, end = var_5668_end_0, end_mask = var_5668_end_mask_0, x = matrix_bd_119_cast_fp16)[name = string("op_5668_cast_fp16")]; - fp16 var_5669_to_fp16 = const()[name = string("op_5669_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_29_cast_fp16 = mul(x = var_5668_cast_fp16, y = var_5669_to_fp16)[name = string("qk_mask_29_cast_fp16")]; - tensor var_5673 = const()[name = string("op_5673"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_29_cast_fp16 = reshape(shape = var_5673, x = query_59_cast_fp16)[name = string("mh_q_29_cast_fp16")]; - fp16 var_5675_to_fp16 = const()[name = string("op_5675_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_5676_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_5675_to_fp16)[name = string("op_5676_cast_fp16")]; - tensor var_5679 = const()[name = string("op_5679"), val = tensor([1, 8, 128, 188])]; - tensor var_5680_cast_fp16 = reshape(shape = var_5679, x = key_29_cast_fp16)[name = string("op_5680_cast_fp16")]; - bool mh_w_57_transpose_x_0 = const()[name = string("mh_w_57_transpose_x_0"), val = bool(true)]; - bool mh_w_57_transpose_y_0 = const()[name = string("mh_w_57_transpose_y_0"), val = bool(false)]; - tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_5676_cast_fp16, y = var_5680_cast_fp16)[name = string("mh_w_57_cast_fp16")]; - tensor mh_w_59_cast_fp16 = add(x = mh_w_57_cast_fp16, y = qk_mask_29_cast_fp16)[name = string("mh_w_59_cast_fp16")]; - tensor var_5684_cast_fp16 = softmax(axis = var_5471, x = mh_w_59_cast_fp16)[name = string("op_5684_cast_fp16")]; - tensor var_5685 = const()[name = string("op_5685"), val = tensor([1, 8, 128, 188])]; - tensor var_5686_cast_fp16 = reshape(shape = var_5685, x = value_29_cast_fp16)[name = string("op_5686_cast_fp16")]; - bool attn_29_transpose_x_0 = const()[name = string("attn_29_transpose_x_0"), val = bool(false)]; - bool attn_29_transpose_y_0 = const()[name = string("attn_29_transpose_y_0"), val = bool(true)]; - tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_5686_cast_fp16, y = var_5684_cast_fp16)[name = string("attn_29_cast_fp16")]; - tensor var_5689 = const()[name = string("op_5689"), val = tensor([1, 1024, 1, 188])]; - tensor input_387_cast_fp16 = reshape(shape = var_5689, x = attn_29_cast_fp16)[name = string("input_387_cast_fp16")]; - string var_5699_pad_type_0 = const()[name = string("op_5699_pad_type_0"), val = string("valid")]; - tensor var_5699_strides_0 = const()[name = string("op_5699_strides_0"), val = tensor([1, 1])]; - tensor var_5699_pad_0 = const()[name = string("op_5699_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5699_dilations_0 = const()[name = string("op_5699_dilations_0"), val = tensor([1, 1])]; - int32 var_5699_groups_0 = const()[name = string("op_5699_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191494144))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191887424))))[name = string("layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5699_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5699_dilations_0, groups = var_5699_groups_0, pad = var_5699_pad_0, pad_type = var_5699_pad_type_0, strides = var_5699_strides_0, weight = layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_387_cast_fp16)[name = string("op_5699_cast_fp16")]; - string var_5705_pad_type_0 = const()[name = string("op_5705_pad_type_0"), val = string("valid")]; - tensor var_5705_strides_0 = const()[name = string("op_5705_strides_0"), val = tensor([1, 1])]; - tensor var_5705_pad_0 = const()[name = string("op_5705_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5705_dilations_0 = const()[name = string("op_5705_dilations_0"), val = tensor([1, 1])]; - int32 var_5705_groups_0 = const()[name = string("op_5705_groups_0"), val = int32(1)]; - tensor layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191897408))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191888512))))[name = string("layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5705_cast_fp16 = conv(dilations = var_5705_dilations_0, groups = var_5705_groups_0, pad = var_5705_pad_0, pad_type = var_5705_pad_type_0, strides = var_5705_strides_0, weight = layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_387_cast_fp16)[name = string("op_5705_cast_fp16")]; - tensor obj_61_cast_fp16 = add(x = var_5699_cast_fp16, y = var_5705_cast_fp16)[name = string("obj_61_cast_fp16")]; - tensor inputs_145_cast_fp16 = add(x = inputs_143_cast_fp16, y = obj_61_cast_fp16)[name = string("inputs_145_cast_fp16")]; - tensor out_145_axes_0 = const()[name = string("out_145_axes_0"), val = tensor([1])]; - fp16 var_5716_to_fp16 = const()[name = string("op_5716_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_145_cast_fp16 = layer_norm(axes = out_145_axes_0, epsilon = var_5716_to_fp16, x = inputs_145_cast_fp16)[name = string("out_145_cast_fp16")]; - tensor input_389_gamma_0_to_fp16 = const()[name = string("input_389_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192028544)))]; - tensor input_389_beta_0_to_fp16 = const()[name = string("input_389_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192030656)))]; - fp16 input_389_epsilon_0_to_fp16 = const()[name = string("input_389_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_389_cast_fp16 = batch_norm(beta = input_389_beta_0_to_fp16, epsilon = input_389_epsilon_0_to_fp16, gamma = input_389_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_145_cast_fp16)[name = string("input_389_cast_fp16")]; - string var_5737_pad_type_0 = const()[name = string("op_5737_pad_type_0"), val = string("valid")]; - tensor var_5737_strides_0 = const()[name = string("op_5737_strides_0"), val = tensor([1, 1])]; - tensor var_5737_pad_0 = const()[name = string("op_5737_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5737_dilations_0 = const()[name = string("op_5737_dilations_0"), val = tensor([1, 1])]; - int32 var_5737_groups_0 = const()[name = string("op_5737_groups_0"), val = int32(1)]; - tensor layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192032768))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192819264))))[name = string("layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5737_cast_fp16 = conv(dilations = var_5737_dilations_0, groups = var_5737_groups_0, pad = var_5737_pad_0, pad_type = var_5737_pad_type_0, strides = var_5737_strides_0, weight = layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_389_cast_fp16)[name = string("op_5737_cast_fp16")]; - string var_5743_pad_type_0 = const()[name = string("op_5743_pad_type_0"), val = string("valid")]; - tensor var_5743_strides_0 = const()[name = string("op_5743_strides_0"), val = tensor([1, 1])]; - tensor var_5743_pad_0 = const()[name = string("op_5743_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5743_dilations_0 = const()[name = string("op_5743_dilations_0"), val = tensor([1, 1])]; - int32 var_5743_groups_0 = const()[name = string("op_5743_groups_0"), val = int32(1)]; - tensor layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192841280))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192821376))))[name = string("layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5743_cast_fp16 = conv(dilations = var_5743_dilations_0, groups = var_5743_groups_0, pad = var_5743_pad_0, pad_type = var_5743_pad_type_0, strides = var_5743_strides_0, weight = layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_389_cast_fp16)[name = string("op_5743_cast_fp16")]; - tensor input_391_cast_fp16 = add(x = var_5737_cast_fp16, y = var_5743_cast_fp16)[name = string("input_391_cast_fp16")]; - int32 input_393_split_num_splits_0 = const()[name = string("input_393_split_num_splits_0"), val = int32(2)]; - int32 input_393_split_axis_0 = const()[name = string("input_393_split_axis_0"), val = int32(1)]; - tensor input_393_split_cast_fp16_0, tensor input_393_split_cast_fp16_1 = split(axis = input_393_split_axis_0, num_splits = input_393_split_num_splits_0, x = input_391_cast_fp16)[name = string("input_393_split_cast_fp16")]; - tensor input_393_split_1_sigmoid_cast_fp16 = sigmoid(x = input_393_split_cast_fp16_1)[name = string("input_393_split_1_sigmoid_cast_fp16")]; - tensor input_393_cast_fp16 = mul(x = input_393_split_cast_fp16_0, y = input_393_split_1_sigmoid_cast_fp16)[name = string("input_393_cast_fp16")]; - string input_395_pad_type_0 = const()[name = string("input_395_pad_type_0"), val = string("custom")]; - tensor input_395_pad_0 = const()[name = string("input_395_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_395_groups_0 = const()[name = string("input_395_groups_0"), val = int32(1024)]; - tensor input_395_strides_0 = const()[name = string("input_395_strides_0"), val = tensor([1, 1])]; - tensor input_395_dilations_0 = const()[name = string("input_395_dilations_0"), val = tensor([1, 1])]; - tensor const_296_to_fp16 = const()[name = string("const_296_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193103488)))]; - tensor const_297_to_fp16 = const()[name = string("const_297_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193121984)))]; - tensor input_397_cast_fp16 = conv(bias = const_297_to_fp16, dilations = input_395_dilations_0, groups = input_395_groups_0, pad = input_395_pad_0, pad_type = input_395_pad_type_0, strides = input_395_strides_0, weight = const_296_to_fp16, x = input_393_cast_fp16)[name = string("input_397_cast_fp16")]; - tensor input_399_cast_fp16 = silu(x = input_397_cast_fp16)[name = string("input_399_cast_fp16")]; - string var_5765_pad_type_0 = const()[name = string("op_5765_pad_type_0"), val = string("valid")]; - tensor var_5765_strides_0 = const()[name = string("op_5765_strides_0"), val = tensor([1, 1])]; - tensor var_5765_pad_0 = const()[name = string("op_5765_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5765_dilations_0 = const()[name = string("op_5765_dilations_0"), val = tensor([1, 1])]; - int32 var_5765_groups_0 = const()[name = string("op_5765_groups_0"), val = int32(1)]; - tensor layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193124096))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193517376))))[name = string("layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5765_cast_fp16 = conv(dilations = var_5765_dilations_0, groups = var_5765_groups_0, pad = var_5765_pad_0, pad_type = var_5765_pad_type_0, strides = var_5765_strides_0, weight = layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_399_cast_fp16)[name = string("op_5765_cast_fp16")]; - string var_5771_pad_type_0 = const()[name = string("op_5771_pad_type_0"), val = string("valid")]; - tensor var_5771_strides_0 = const()[name = string("op_5771_strides_0"), val = tensor([1, 1])]; - tensor var_5771_pad_0 = const()[name = string("op_5771_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5771_dilations_0 = const()[name = string("op_5771_dilations_0"), val = tensor([1, 1])]; - int32 var_5771_groups_0 = const()[name = string("op_5771_groups_0"), val = int32(1)]; - tensor layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193526784))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193518464))))[name = string("layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5771_cast_fp16 = conv(dilations = var_5771_dilations_0, groups = var_5771_groups_0, pad = var_5771_pad_0, pad_type = var_5771_pad_type_0, strides = var_5771_strides_0, weight = layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_399_cast_fp16)[name = string("op_5771_cast_fp16")]; - tensor x_89_cast_fp16 = add(x = var_5765_cast_fp16, y = var_5771_cast_fp16)[name = string("x_89_cast_fp16")]; - tensor inputs_147_cast_fp16 = add(x = inputs_145_cast_fp16, y = x_89_cast_fp16)[name = string("inputs_147_cast_fp16")]; - tensor out_147_axes_0 = const()[name = string("out_147_axes_0"), val = tensor([1])]; - fp16 var_5782_to_fp16 = const()[name = string("op_5782_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_147_cast_fp16 = layer_norm(axes = out_147_axes_0, epsilon = var_5782_to_fp16, x = inputs_147_cast_fp16)[name = string("out_147_cast_fp16")]; - tensor input_401_gamma_0_to_fp16 = const()[name = string("input_401_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193657920)))]; - tensor input_401_beta_0_to_fp16 = const()[name = string("input_401_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193660032)))]; - fp16 input_401_epsilon_0_to_fp16 = const()[name = string("input_401_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_401_cast_fp16 = batch_norm(beta = input_401_beta_0_to_fp16, epsilon = input_401_epsilon_0_to_fp16, gamma = input_401_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_147_cast_fp16)[name = string("input_401_cast_fp16")]; - string var_5802_pad_type_0 = const()[name = string("op_5802_pad_type_0"), val = string("valid")]; - tensor var_5802_strides_0 = const()[name = string("op_5802_strides_0"), val = tensor([1, 1])]; - tensor var_5802_pad_0 = const()[name = string("op_5802_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5802_dilations_0 = const()[name = string("op_5802_dilations_0"), val = tensor([1, 1])]; - int32 var_5802_groups_0 = const()[name = string("op_5802_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193662144))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195235072))))[name = string("layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5802_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5802_dilations_0, groups = var_5802_groups_0, pad = var_5802_pad_0, pad_type = var_5802_pad_type_0, strides = var_5802_strides_0, weight = layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_401_cast_fp16)[name = string("op_5802_cast_fp16")]; - string var_5808_pad_type_0 = const()[name = string("op_5808_pad_type_0"), val = string("valid")]; - tensor var_5808_strides_0 = const()[name = string("op_5808_strides_0"), val = tensor([1, 1])]; - tensor var_5808_pad_0 = const()[name = string("op_5808_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5808_dilations_0 = const()[name = string("op_5808_dilations_0"), val = tensor([1, 1])]; - int32 var_5808_groups_0 = const()[name = string("op_5808_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195282688))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195239232))))[name = string("layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5808_cast_fp16 = conv(dilations = var_5808_dilations_0, groups = var_5808_groups_0, pad = var_5808_pad_0, pad_type = var_5808_pad_type_0, strides = var_5808_strides_0, weight = layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_401_cast_fp16)[name = string("op_5808_cast_fp16")]; - tensor input_403_cast_fp16 = add(x = var_5802_cast_fp16, y = var_5808_cast_fp16)[name = string("input_403_cast_fp16")]; - tensor input_405_cast_fp16 = silu(x = input_403_cast_fp16)[name = string("input_405_cast_fp16")]; - string var_5819_pad_type_0 = const()[name = string("op_5819_pad_type_0"), val = string("valid")]; - tensor var_5819_strides_0 = const()[name = string("op_5819_strides_0"), val = tensor([1, 1])]; - tensor var_5819_pad_0 = const()[name = string("op_5819_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5819_dilations_0 = const()[name = string("op_5819_dilations_0"), val = tensor([1, 1])]; - int32 var_5819_groups_0 = const()[name = string("op_5819_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195807040))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197379968))))[name = string("layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5819_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5819_dilations_0, groups = var_5819_groups_0, pad = var_5819_pad_0, pad_type = var_5819_pad_type_0, strides = var_5819_strides_0, weight = layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_405_cast_fp16)[name = string("op_5819_cast_fp16")]; - string var_5825_pad_type_0 = const()[name = string("op_5825_pad_type_0"), val = string("valid")]; - tensor var_5825_strides_0 = const()[name = string("op_5825_strides_0"), val = tensor([1, 1])]; - tensor var_5825_pad_0 = const()[name = string("op_5825_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5825_dilations_0 = const()[name = string("op_5825_dilations_0"), val = tensor([1, 1])]; - int32 var_5825_groups_0 = const()[name = string("op_5825_groups_0"), val = int32(1)]; - tensor layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197437440))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197381056))))[name = string("layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5825_cast_fp16 = conv(dilations = var_5825_dilations_0, groups = var_5825_groups_0, pad = var_5825_pad_0, pad_type = var_5825_pad_type_0, strides = var_5825_strides_0, weight = layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_405_cast_fp16)[name = string("op_5825_cast_fp16")]; - tensor x_91_cast_fp16 = add(x = var_5819_cast_fp16, y = var_5825_cast_fp16)[name = string("x_91_cast_fp16")]; - fp16 var_5827_to_fp16 = const()[name = string("op_5827_to_fp16"), val = fp16(0x1p-1)]; - tensor var_5828_cast_fp16 = mul(x = x_91_cast_fp16, y = var_5827_to_fp16)[name = string("op_5828_cast_fp16")]; - tensor inputs_149_cast_fp16 = add(x = inputs_147_cast_fp16, y = var_5828_cast_fp16)[name = string("inputs_149_cast_fp16")]; - tensor out_149_axes_0 = const()[name = string("out_149_axes_0"), val = tensor([1])]; - fp16 var_5838_to_fp16 = const()[name = string("op_5838_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_149_cast_fp16 = layer_norm(axes = out_149_axes_0, epsilon = var_5838_to_fp16, x = inputs_149_cast_fp16)[name = string("out_149_cast_fp16")]; - tensor inputs_151_gamma_0_to_fp16 = const()[name = string("inputs_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197961792)))]; - tensor inputs_151_beta_0_to_fp16 = const()[name = string("inputs_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197963904)))]; - fp16 inputs_151_epsilon_0_to_fp16 = const()[name = string("inputs_151_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_151_cast_fp16 = batch_norm(beta = inputs_151_beta_0_to_fp16, epsilon = inputs_151_epsilon_0_to_fp16, gamma = inputs_151_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_149_cast_fp16)[name = string("inputs_151_cast_fp16")]; - int32 var_5852 = const()[name = string("op_5852"), val = int32(3)]; - tensor out_151_axes_0 = const()[name = string("out_151_axes_0"), val = tensor([1])]; - fp16 var_5883_to_fp16 = const()[name = string("op_5883_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_151_cast_fp16 = layer_norm(axes = out_151_axes_0, epsilon = var_5883_to_fp16, x = inputs_151_cast_fp16)[name = string("out_151_cast_fp16")]; - tensor input_407_gamma_0_to_fp16 = const()[name = string("input_407_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197966016)))]; - tensor input_407_beta_0_to_fp16 = const()[name = string("input_407_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197968128)))]; - fp16 input_407_epsilon_0_to_fp16 = const()[name = string("input_407_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_407_cast_fp16 = batch_norm(beta = input_407_beta_0_to_fp16, epsilon = input_407_epsilon_0_to_fp16, gamma = input_407_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_151_cast_fp16)[name = string("input_407_cast_fp16")]; - string var_5903_pad_type_0 = const()[name = string("op_5903_pad_type_0"), val = string("valid")]; - tensor var_5903_strides_0 = const()[name = string("op_5903_strides_0"), val = tensor([1, 1])]; - tensor var_5903_pad_0 = const()[name = string("op_5903_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5903_dilations_0 = const()[name = string("op_5903_dilations_0"), val = tensor([1, 1])]; - int32 var_5903_groups_0 = const()[name = string("op_5903_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197970240))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199543168))))[name = string("layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_5903_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5903_dilations_0, groups = var_5903_groups_0, pad = var_5903_pad_0, pad_type = var_5903_pad_type_0, strides = var_5903_strides_0, weight = layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_407_cast_fp16)[name = string("op_5903_cast_fp16")]; - string var_5909_pad_type_0 = const()[name = string("op_5909_pad_type_0"), val = string("valid")]; - tensor var_5909_strides_0 = const()[name = string("op_5909_strides_0"), val = tensor([1, 1])]; - tensor var_5909_pad_0 = const()[name = string("op_5909_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5909_dilations_0 = const()[name = string("op_5909_dilations_0"), val = tensor([1, 1])]; - int32 var_5909_groups_0 = const()[name = string("op_5909_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199597376))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199547328))))[name = string("layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5909_cast_fp16 = conv(dilations = var_5909_dilations_0, groups = var_5909_groups_0, pad = var_5909_pad_0, pad_type = var_5909_pad_type_0, strides = var_5909_strides_0, weight = layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_407_cast_fp16)[name = string("op_5909_cast_fp16")]; - tensor input_409_cast_fp16 = add(x = var_5903_cast_fp16, y = var_5909_cast_fp16)[name = string("input_409_cast_fp16")]; - tensor input_411_cast_fp16 = silu(x = input_409_cast_fp16)[name = string("input_411_cast_fp16")]; - string var_5920_pad_type_0 = const()[name = string("op_5920_pad_type_0"), val = string("valid")]; - tensor var_5920_strides_0 = const()[name = string("op_5920_strides_0"), val = tensor([1, 1])]; - tensor var_5920_pad_0 = const()[name = string("op_5920_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5920_dilations_0 = const()[name = string("op_5920_dilations_0"), val = tensor([1, 1])]; - int32 var_5920_groups_0 = const()[name = string("op_5920_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200121728))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201694656))))[name = string("layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_5920_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5920_dilations_0, groups = var_5920_groups_0, pad = var_5920_pad_0, pad_type = var_5920_pad_type_0, strides = var_5920_strides_0, weight = layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_411_cast_fp16)[name = string("op_5920_cast_fp16")]; - string var_5926_pad_type_0 = const()[name = string("op_5926_pad_type_0"), val = string("valid")]; - tensor var_5926_strides_0 = const()[name = string("op_5926_strides_0"), val = tensor([1, 1])]; - tensor var_5926_pad_0 = const()[name = string("op_5926_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5926_dilations_0 = const()[name = string("op_5926_dilations_0"), val = tensor([1, 1])]; - int32 var_5926_groups_0 = const()[name = string("op_5926_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201763328))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201695744))))[name = string("layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5926_cast_fp16 = conv(dilations = var_5926_dilations_0, groups = var_5926_groups_0, pad = var_5926_pad_0, pad_type = var_5926_pad_type_0, strides = var_5926_strides_0, weight = layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_411_cast_fp16)[name = string("op_5926_cast_fp16")]; - tensor x_93_cast_fp16 = add(x = var_5920_cast_fp16, y = var_5926_cast_fp16)[name = string("x_93_cast_fp16")]; - fp16 var_5928_to_fp16 = const()[name = string("op_5928_to_fp16"), val = fp16(0x1p-1)]; - tensor var_5929_cast_fp16 = mul(x = x_93_cast_fp16, y = var_5928_to_fp16)[name = string("op_5929_cast_fp16")]; - tensor inputs_153_cast_fp16 = add(x = inputs_151_cast_fp16, y = var_5929_cast_fp16)[name = string("inputs_153_cast_fp16")]; - tensor out_153_axes_0 = const()[name = string("out_153_axes_0"), val = tensor([1])]; - fp16 var_5939_to_fp16 = const()[name = string("op_5939_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_153_cast_fp16 = layer_norm(axes = out_153_axes_0, epsilon = var_5939_to_fp16, x = inputs_153_cast_fp16)[name = string("out_153_cast_fp16")]; - tensor obj_63_gamma_0_to_fp16 = const()[name = string("obj_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202287680)))]; - tensor obj_63_beta_0_to_fp16 = const()[name = string("obj_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202289792)))]; - fp16 obj_63_epsilon_0_to_fp16 = const()[name = string("obj_63_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_63_cast_fp16 = batch_norm(beta = obj_63_beta_0_to_fp16, epsilon = obj_63_epsilon_0_to_fp16, gamma = obj_63_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_153_cast_fp16)[name = string("obj_63_cast_fp16")]; - string var_5964_pad_type_0 = const()[name = string("op_5964_pad_type_0"), val = string("valid")]; - tensor var_5964_strides_0 = const()[name = string("op_5964_strides_0"), val = tensor([1, 1])]; - tensor var_5964_pad_0 = const()[name = string("op_5964_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5964_dilations_0 = const()[name = string("op_5964_dilations_0"), val = tensor([1, 1])]; - int32 var_5964_groups_0 = const()[name = string("op_5964_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202291904))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202685184))))[name = string("layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5964_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5964_dilations_0, groups = var_5964_groups_0, pad = var_5964_pad_0, pad_type = var_5964_pad_type_0, strides = var_5964_strides_0, weight = layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = string("op_5964_cast_fp16")]; - string var_5970_pad_type_0 = const()[name = string("op_5970_pad_type_0"), val = string("valid")]; - tensor var_5970_strides_0 = const()[name = string("op_5970_strides_0"), val = tensor([1, 1])]; - tensor var_5970_pad_0 = const()[name = string("op_5970_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5970_dilations_0 = const()[name = string("op_5970_dilations_0"), val = tensor([1, 1])]; - int32 var_5970_groups_0 = const()[name = string("op_5970_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202696000))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202686272))))[name = string("layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5970_cast_fp16 = conv(dilations = var_5970_dilations_0, groups = var_5970_groups_0, pad = var_5970_pad_0, pad_type = var_5970_pad_type_0, strides = var_5970_strides_0, weight = layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = string("op_5970_cast_fp16")]; - tensor query_61_cast_fp16 = add(x = var_5964_cast_fp16, y = var_5970_cast_fp16)[name = string("query_61_cast_fp16")]; - string var_5979_pad_type_0 = const()[name = string("op_5979_pad_type_0"), val = string("valid")]; - tensor var_5979_strides_0 = const()[name = string("op_5979_strides_0"), val = tensor([1, 1])]; - tensor var_5979_pad_0 = const()[name = string("op_5979_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5979_dilations_0 = const()[name = string("op_5979_dilations_0"), val = tensor([1, 1])]; - int32 var_5979_groups_0 = const()[name = string("op_5979_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202827136))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203220416))))[name = string("layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5979_cast_fp16 = conv(dilations = var_5979_dilations_0, groups = var_5979_groups_0, pad = var_5979_pad_0, pad_type = var_5979_pad_type_0, strides = var_5979_strides_0, weight = layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = string("op_5979_cast_fp16")]; - string var_5985_pad_type_0 = const()[name = string("op_5985_pad_type_0"), val = string("valid")]; - tensor var_5985_strides_0 = const()[name = string("op_5985_strides_0"), val = tensor([1, 1])]; - tensor var_5985_pad_0 = const()[name = string("op_5985_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5985_dilations_0 = const()[name = string("op_5985_dilations_0"), val = tensor([1, 1])]; - int32 var_5985_groups_0 = const()[name = string("op_5985_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203232640))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203221504))))[name = string("layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_5985_cast_fp16 = conv(dilations = var_5985_dilations_0, groups = var_5985_groups_0, pad = var_5985_pad_0, pad_type = var_5985_pad_type_0, strides = var_5985_strides_0, weight = layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = string("op_5985_cast_fp16")]; - tensor key_31_cast_fp16 = add(x = var_5979_cast_fp16, y = var_5985_cast_fp16)[name = string("key_31_cast_fp16")]; - string var_5995_pad_type_0 = const()[name = string("op_5995_pad_type_0"), val = string("valid")]; - tensor var_5995_strides_0 = const()[name = string("op_5995_strides_0"), val = tensor([1, 1])]; - tensor var_5995_pad_0 = const()[name = string("op_5995_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_5995_dilations_0 = const()[name = string("op_5995_dilations_0"), val = tensor([1, 1])]; - int32 var_5995_groups_0 = const()[name = string("op_5995_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203363776))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203757056))))[name = string("layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_5995_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5995_dilations_0, groups = var_5995_groups_0, pad = var_5995_pad_0, pad_type = var_5995_pad_type_0, strides = var_5995_strides_0, weight = layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = string("op_5995_cast_fp16")]; - string var_6001_pad_type_0 = const()[name = string("op_6001_pad_type_0"), val = string("valid")]; - tensor var_6001_strides_0 = const()[name = string("op_6001_strides_0"), val = tensor([1, 1])]; - tensor var_6001_pad_0 = const()[name = string("op_6001_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6001_dilations_0 = const()[name = string("op_6001_dilations_0"), val = tensor([1, 1])]; - int32 var_6001_groups_0 = const()[name = string("op_6001_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203765760))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203758144))))[name = string("layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6001_cast_fp16 = conv(dilations = var_6001_dilations_0, groups = var_6001_groups_0, pad = var_6001_pad_0, pad_type = var_6001_pad_type_0, strides = var_6001_strides_0, weight = layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = string("op_6001_cast_fp16")]; - tensor value_31_cast_fp16 = add(x = var_5995_cast_fp16, y = var_6001_cast_fp16)[name = string("value_31_cast_fp16")]; - tensor var_6004_to_fp16 = const()[name = string("op_6004_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203896896)))]; - tensor query_63_cast_fp16 = add(x = query_61_cast_fp16, y = var_6004_to_fp16)[name = string("query_63_cast_fp16")]; - tensor var_6007_to_fp16 = const()[name = string("op_6007_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203899008)))]; - tensor q_with_bias_v_31_cast_fp16 = add(x = query_61_cast_fp16, y = var_6007_to_fp16)[name = string("q_with_bias_v_31_cast_fp16")]; - string var_6017_pad_type_0 = const()[name = string("op_6017_pad_type_0"), val = string("valid")]; - tensor var_6017_strides_0 = const()[name = string("op_6017_strides_0"), val = tensor([1, 1])]; - tensor var_6017_pad_0 = const()[name = string("op_6017_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6017_dilations_0 = const()[name = string("op_6017_dilations_0"), val = tensor([1, 1])]; - int32 var_6017_groups_0 = const()[name = string("op_6017_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203901120))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204294400))))[name = string("layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_6017_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6017_dilations_0, groups = var_6017_groups_0, pad = var_6017_pad_0, pad_type = var_6017_pad_type_0, strides = var_6017_strides_0, weight = layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_6017_cast_fp16")]; - string var_6023_pad_type_0 = const()[name = string("op_6023_pad_type_0"), val = string("valid")]; - tensor var_6023_strides_0 = const()[name = string("op_6023_strides_0"), val = tensor([1, 1])]; - tensor var_6023_pad_0 = const()[name = string("op_6023_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6023_dilations_0 = const()[name = string("op_6023_dilations_0"), val = tensor([1, 1])]; - int32 var_6023_groups_0 = const()[name = string("op_6023_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204333504))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204295488))))[name = string("layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6023_cast_fp16 = conv(dilations = var_6023_dilations_0, groups = var_6023_groups_0, pad = var_6023_pad_0, pad_type = var_6023_pad_type_0, strides = var_6023_strides_0, weight = layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_6023_cast_fp16")]; - tensor p_31_cast_fp16 = add(x = var_6017_cast_fp16, y = var_6023_cast_fp16)[name = string("p_31_cast_fp16")]; - tensor var_6027 = const()[name = string("op_6027"), val = tensor([1, 8, 128, 188])]; - tensor var_6028_cast_fp16 = reshape(shape = var_6027, x = q_with_bias_v_31_cast_fp16)[name = string("op_6028_cast_fp16")]; - tensor var_6029 = const()[name = string("op_6029"), val = tensor([1, 8, 128, -1])]; - tensor var_6030_cast_fp16 = reshape(shape = var_6029, x = p_31_cast_fp16)[name = string("op_6030_cast_fp16")]; - bool matrix_bd_121_transpose_x_0 = const()[name = string("matrix_bd_121_transpose_x_0"), val = bool(true)]; - bool matrix_bd_121_transpose_y_0 = const()[name = string("matrix_bd_121_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_121_cast_fp16 = matmul(transpose_x = matrix_bd_121_transpose_x_0, transpose_y = matrix_bd_121_transpose_y_0, x = var_6028_cast_fp16, y = var_6030_cast_fp16)[name = string("matrix_bd_121_cast_fp16")]; - tensor matrix_bd_123_pad_0 = const()[name = string("matrix_bd_123_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_123_mode_0 = const()[name = string("matrix_bd_123_mode_0"), val = string("constant")]; - fp16 const_175_to_fp16 = const()[name = string("const_175_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_123_cast_fp16 = pad(constant_val = const_175_to_fp16, mode = matrix_bd_123_mode_0, pad = matrix_bd_123_pad_0, x = matrix_bd_121_cast_fp16)[name = string("matrix_bd_123_cast_fp16")]; - tensor var_6039 = const()[name = string("op_6039"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_125_cast_fp16 = reshape(shape = var_6039, x = matrix_bd_123_cast_fp16)[name = string("matrix_bd_125_cast_fp16")]; - tensor var_6043_begin_0 = const()[name = string("op_6043_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_6043_end_0 = const()[name = string("op_6043_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_6043_end_mask_0 = const()[name = string("op_6043_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_6043_cast_fp16 = slice_by_index(begin = var_6043_begin_0, end = var_6043_end_0, end_mask = var_6043_end_mask_0, x = matrix_bd_125_cast_fp16)[name = string("op_6043_cast_fp16")]; - tensor var_6044 = const()[name = string("op_6044"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_127_cast_fp16 = reshape(shape = var_6044, x = var_6043_cast_fp16)[name = string("matrix_bd_127_cast_fp16")]; - tensor var_6049_begin_0 = const()[name = string("op_6049_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6049_end_0 = const()[name = string("op_6049_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_6049_end_mask_0 = const()[name = string("op_6049_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_6049_cast_fp16 = slice_by_index(begin = var_6049_begin_0, end = var_6049_end_0, end_mask = var_6049_end_mask_0, x = matrix_bd_127_cast_fp16)[name = string("op_6049_cast_fp16")]; - fp16 var_6050_to_fp16 = const()[name = string("op_6050_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_31_cast_fp16 = mul(x = var_6049_cast_fp16, y = var_6050_to_fp16)[name = string("qk_mask_31_cast_fp16")]; - tensor var_6054 = const()[name = string("op_6054"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_31_cast_fp16 = reshape(shape = var_6054, x = query_63_cast_fp16)[name = string("mh_q_31_cast_fp16")]; - fp16 var_6056_to_fp16 = const()[name = string("op_6056_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_6057_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_6056_to_fp16)[name = string("op_6057_cast_fp16")]; - tensor var_6060 = const()[name = string("op_6060"), val = tensor([1, 8, 128, 188])]; - tensor var_6061_cast_fp16 = reshape(shape = var_6060, x = key_31_cast_fp16)[name = string("op_6061_cast_fp16")]; - bool mh_w_61_transpose_x_0 = const()[name = string("mh_w_61_transpose_x_0"), val = bool(true)]; - bool mh_w_61_transpose_y_0 = const()[name = string("mh_w_61_transpose_y_0"), val = bool(false)]; - tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_6057_cast_fp16, y = var_6061_cast_fp16)[name = string("mh_w_61_cast_fp16")]; - tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = qk_mask_31_cast_fp16)[name = string("mh_w_63_cast_fp16")]; - tensor var_6065_cast_fp16 = softmax(axis = var_5852, x = mh_w_63_cast_fp16)[name = string("op_6065_cast_fp16")]; - tensor var_6066 = const()[name = string("op_6066"), val = tensor([1, 8, 128, 188])]; - tensor var_6067_cast_fp16 = reshape(shape = var_6066, x = value_31_cast_fp16)[name = string("op_6067_cast_fp16")]; - bool attn_31_transpose_x_0 = const()[name = string("attn_31_transpose_x_0"), val = bool(false)]; - bool attn_31_transpose_y_0 = const()[name = string("attn_31_transpose_y_0"), val = bool(true)]; - tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_6067_cast_fp16, y = var_6065_cast_fp16)[name = string("attn_31_cast_fp16")]; - tensor var_6070 = const()[name = string("op_6070"), val = tensor([1, 1024, 1, 188])]; - tensor input_413_cast_fp16 = reshape(shape = var_6070, x = attn_31_cast_fp16)[name = string("input_413_cast_fp16")]; - string var_6080_pad_type_0 = const()[name = string("op_6080_pad_type_0"), val = string("valid")]; - tensor var_6080_strides_0 = const()[name = string("op_6080_strides_0"), val = tensor([1, 1])]; - tensor var_6080_pad_0 = const()[name = string("op_6080_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6080_dilations_0 = const()[name = string("op_6080_dilations_0"), val = tensor([1, 1])]; - int32 var_6080_groups_0 = const()[name = string("op_6080_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204464640))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204857920))))[name = string("layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6080_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6080_dilations_0, groups = var_6080_groups_0, pad = var_6080_pad_0, pad_type = var_6080_pad_type_0, strides = var_6080_strides_0, weight = layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_413_cast_fp16)[name = string("op_6080_cast_fp16")]; - string var_6086_pad_type_0 = const()[name = string("op_6086_pad_type_0"), val = string("valid")]; - tensor var_6086_strides_0 = const()[name = string("op_6086_strides_0"), val = tensor([1, 1])]; - tensor var_6086_pad_0 = const()[name = string("op_6086_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6086_dilations_0 = const()[name = string("op_6086_dilations_0"), val = tensor([1, 1])]; - int32 var_6086_groups_0 = const()[name = string("op_6086_groups_0"), val = int32(1)]; - tensor layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204866944))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204859008))))[name = string("layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6086_cast_fp16 = conv(dilations = var_6086_dilations_0, groups = var_6086_groups_0, pad = var_6086_pad_0, pad_type = var_6086_pad_type_0, strides = var_6086_strides_0, weight = layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_413_cast_fp16)[name = string("op_6086_cast_fp16")]; - tensor obj_65_cast_fp16 = add(x = var_6080_cast_fp16, y = var_6086_cast_fp16)[name = string("obj_65_cast_fp16")]; - tensor inputs_155_cast_fp16 = add(x = inputs_153_cast_fp16, y = obj_65_cast_fp16)[name = string("inputs_155_cast_fp16")]; - tensor out_155_axes_0 = const()[name = string("out_155_axes_0"), val = tensor([1])]; - fp16 var_6097_to_fp16 = const()[name = string("op_6097_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_155_cast_fp16 = layer_norm(axes = out_155_axes_0, epsilon = var_6097_to_fp16, x = inputs_155_cast_fp16)[name = string("out_155_cast_fp16")]; - tensor input_415_gamma_0_to_fp16 = const()[name = string("input_415_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204998080)))]; - tensor input_415_beta_0_to_fp16 = const()[name = string("input_415_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205000192)))]; - fp16 input_415_epsilon_0_to_fp16 = const()[name = string("input_415_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_415_cast_fp16 = batch_norm(beta = input_415_beta_0_to_fp16, epsilon = input_415_epsilon_0_to_fp16, gamma = input_415_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_155_cast_fp16)[name = string("input_415_cast_fp16")]; - string var_6118_pad_type_0 = const()[name = string("op_6118_pad_type_0"), val = string("valid")]; - tensor var_6118_strides_0 = const()[name = string("op_6118_strides_0"), val = tensor([1, 1])]; - tensor var_6118_pad_0 = const()[name = string("op_6118_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6118_dilations_0 = const()[name = string("op_6118_dilations_0"), val = tensor([1, 1])]; - int32 var_6118_groups_0 = const()[name = string("op_6118_groups_0"), val = int32(1)]; - tensor layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205002304))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205788800))))[name = string("layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6118_cast_fp16 = conv(dilations = var_6118_dilations_0, groups = var_6118_groups_0, pad = var_6118_pad_0, pad_type = var_6118_pad_type_0, strides = var_6118_strides_0, weight = layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_415_cast_fp16)[name = string("op_6118_cast_fp16")]; - string var_6124_pad_type_0 = const()[name = string("op_6124_pad_type_0"), val = string("valid")]; - tensor var_6124_strides_0 = const()[name = string("op_6124_strides_0"), val = tensor([1, 1])]; - tensor var_6124_pad_0 = const()[name = string("op_6124_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6124_dilations_0 = const()[name = string("op_6124_dilations_0"), val = tensor([1, 1])]; - int32 var_6124_groups_0 = const()[name = string("op_6124_groups_0"), val = int32(1)]; - tensor layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205810368))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205790912))))[name = string("layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6124_cast_fp16 = conv(dilations = var_6124_dilations_0, groups = var_6124_groups_0, pad = var_6124_pad_0, pad_type = var_6124_pad_type_0, strides = var_6124_strides_0, weight = layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_415_cast_fp16)[name = string("op_6124_cast_fp16")]; - tensor input_417_cast_fp16 = add(x = var_6118_cast_fp16, y = var_6124_cast_fp16)[name = string("input_417_cast_fp16")]; - int32 input_419_split_num_splits_0 = const()[name = string("input_419_split_num_splits_0"), val = int32(2)]; - int32 input_419_split_axis_0 = const()[name = string("input_419_split_axis_0"), val = int32(1)]; - tensor input_419_split_cast_fp16_0, tensor input_419_split_cast_fp16_1 = split(axis = input_419_split_axis_0, num_splits = input_419_split_num_splits_0, x = input_417_cast_fp16)[name = string("input_419_split_cast_fp16")]; - tensor input_419_split_1_sigmoid_cast_fp16 = sigmoid(x = input_419_split_cast_fp16_1)[name = string("input_419_split_1_sigmoid_cast_fp16")]; - tensor input_419_cast_fp16 = mul(x = input_419_split_cast_fp16_0, y = input_419_split_1_sigmoid_cast_fp16)[name = string("input_419_cast_fp16")]; - string input_421_pad_type_0 = const()[name = string("input_421_pad_type_0"), val = string("custom")]; - tensor input_421_pad_0 = const()[name = string("input_421_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_421_groups_0 = const()[name = string("input_421_groups_0"), val = int32(1024)]; - tensor input_421_strides_0 = const()[name = string("input_421_strides_0"), val = tensor([1, 1])]; - tensor input_421_dilations_0 = const()[name = string("input_421_dilations_0"), val = tensor([1, 1])]; - tensor const_298_to_fp16 = const()[name = string("const_298_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206072576)))]; - tensor const_299_to_fp16 = const()[name = string("const_299_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206091072)))]; - tensor input_423_cast_fp16 = conv(bias = const_299_to_fp16, dilations = input_421_dilations_0, groups = input_421_groups_0, pad = input_421_pad_0, pad_type = input_421_pad_type_0, strides = input_421_strides_0, weight = const_298_to_fp16, x = input_419_cast_fp16)[name = string("input_423_cast_fp16")]; - tensor input_425_cast_fp16 = silu(x = input_423_cast_fp16)[name = string("input_425_cast_fp16")]; - string var_6146_pad_type_0 = const()[name = string("op_6146_pad_type_0"), val = string("valid")]; - tensor var_6146_strides_0 = const()[name = string("op_6146_strides_0"), val = tensor([1, 1])]; - tensor var_6146_pad_0 = const()[name = string("op_6146_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6146_dilations_0 = const()[name = string("op_6146_dilations_0"), val = tensor([1, 1])]; - int32 var_6146_groups_0 = const()[name = string("op_6146_groups_0"), val = int32(1)]; - tensor layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206093184))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206486464))))[name = string("layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6146_cast_fp16 = conv(dilations = var_6146_dilations_0, groups = var_6146_groups_0, pad = var_6146_pad_0, pad_type = var_6146_pad_type_0, strides = var_6146_strides_0, weight = layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_425_cast_fp16)[name = string("op_6146_cast_fp16")]; - string var_6152_pad_type_0 = const()[name = string("op_6152_pad_type_0"), val = string("valid")]; - tensor var_6152_strides_0 = const()[name = string("op_6152_strides_0"), val = tensor([1, 1])]; - tensor var_6152_pad_0 = const()[name = string("op_6152_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6152_dilations_0 = const()[name = string("op_6152_dilations_0"), val = tensor([1, 1])]; - int32 var_6152_groups_0 = const()[name = string("op_6152_groups_0"), val = int32(1)]; - tensor layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206495872))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206487552))))[name = string("layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6152_cast_fp16 = conv(dilations = var_6152_dilations_0, groups = var_6152_groups_0, pad = var_6152_pad_0, pad_type = var_6152_pad_type_0, strides = var_6152_strides_0, weight = layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_425_cast_fp16)[name = string("op_6152_cast_fp16")]; - tensor x_95_cast_fp16 = add(x = var_6146_cast_fp16, y = var_6152_cast_fp16)[name = string("x_95_cast_fp16")]; - tensor inputs_157_cast_fp16 = add(x = inputs_155_cast_fp16, y = x_95_cast_fp16)[name = string("inputs_157_cast_fp16")]; - tensor out_157_axes_0 = const()[name = string("out_157_axes_0"), val = tensor([1])]; - fp16 var_6163_to_fp16 = const()[name = string("op_6163_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_157_cast_fp16 = layer_norm(axes = out_157_axes_0, epsilon = var_6163_to_fp16, x = inputs_157_cast_fp16)[name = string("out_157_cast_fp16")]; - tensor input_427_gamma_0_to_fp16 = const()[name = string("input_427_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206627008)))]; - tensor input_427_beta_0_to_fp16 = const()[name = string("input_427_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206629120)))]; - fp16 input_427_epsilon_0_to_fp16 = const()[name = string("input_427_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_427_cast_fp16 = batch_norm(beta = input_427_beta_0_to_fp16, epsilon = input_427_epsilon_0_to_fp16, gamma = input_427_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_157_cast_fp16)[name = string("input_427_cast_fp16")]; - string var_6183_pad_type_0 = const()[name = string("op_6183_pad_type_0"), val = string("valid")]; - tensor var_6183_strides_0 = const()[name = string("op_6183_strides_0"), val = tensor([1, 1])]; - tensor var_6183_pad_0 = const()[name = string("op_6183_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6183_dilations_0 = const()[name = string("op_6183_dilations_0"), val = tensor([1, 1])]; - int32 var_6183_groups_0 = const()[name = string("op_6183_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206631232))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208204160))))[name = string("layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6183_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6183_dilations_0, groups = var_6183_groups_0, pad = var_6183_pad_0, pad_type = var_6183_pad_type_0, strides = var_6183_strides_0, weight = layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_427_cast_fp16)[name = string("op_6183_cast_fp16")]; - string var_6189_pad_type_0 = const()[name = string("op_6189_pad_type_0"), val = string("valid")]; - tensor var_6189_strides_0 = const()[name = string("op_6189_strides_0"), val = tensor([1, 1])]; - tensor var_6189_pad_0 = const()[name = string("op_6189_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6189_dilations_0 = const()[name = string("op_6189_dilations_0"), val = tensor([1, 1])]; - int32 var_6189_groups_0 = const()[name = string("op_6189_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208250368))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208208320))))[name = string("layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6189_cast_fp16 = conv(dilations = var_6189_dilations_0, groups = var_6189_groups_0, pad = var_6189_pad_0, pad_type = var_6189_pad_type_0, strides = var_6189_strides_0, weight = layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_427_cast_fp16)[name = string("op_6189_cast_fp16")]; - tensor input_429_cast_fp16 = add(x = var_6183_cast_fp16, y = var_6189_cast_fp16)[name = string("input_429_cast_fp16")]; - tensor input_431_cast_fp16 = silu(x = input_429_cast_fp16)[name = string("input_431_cast_fp16")]; - string var_6200_pad_type_0 = const()[name = string("op_6200_pad_type_0"), val = string("valid")]; - tensor var_6200_strides_0 = const()[name = string("op_6200_strides_0"), val = tensor([1, 1])]; - tensor var_6200_pad_0 = const()[name = string("op_6200_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6200_dilations_0 = const()[name = string("op_6200_dilations_0"), val = tensor([1, 1])]; - int32 var_6200_groups_0 = const()[name = string("op_6200_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208774720))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210347648))))[name = string("layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6200_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6200_dilations_0, groups = var_6200_groups_0, pad = var_6200_pad_0, pad_type = var_6200_pad_type_0, strides = var_6200_strides_0, weight = layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_431_cast_fp16)[name = string("op_6200_cast_fp16")]; - string var_6206_pad_type_0 = const()[name = string("op_6206_pad_type_0"), val = string("valid")]; - tensor var_6206_strides_0 = const()[name = string("op_6206_strides_0"), val = tensor([1, 1])]; - tensor var_6206_pad_0 = const()[name = string("op_6206_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6206_dilations_0 = const()[name = string("op_6206_dilations_0"), val = tensor([1, 1])]; - int32 var_6206_groups_0 = const()[name = string("op_6206_groups_0"), val = int32(1)]; - tensor layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210403968))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210348736))))[name = string("layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6206_cast_fp16 = conv(dilations = var_6206_dilations_0, groups = var_6206_groups_0, pad = var_6206_pad_0, pad_type = var_6206_pad_type_0, strides = var_6206_strides_0, weight = layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_431_cast_fp16)[name = string("op_6206_cast_fp16")]; - tensor x_97_cast_fp16 = add(x = var_6200_cast_fp16, y = var_6206_cast_fp16)[name = string("x_97_cast_fp16")]; - fp16 var_6208_to_fp16 = const()[name = string("op_6208_to_fp16"), val = fp16(0x1p-1)]; - tensor var_6209_cast_fp16 = mul(x = x_97_cast_fp16, y = var_6208_to_fp16)[name = string("op_6209_cast_fp16")]; - tensor inputs_159_cast_fp16 = add(x = inputs_157_cast_fp16, y = var_6209_cast_fp16)[name = string("inputs_159_cast_fp16")]; - tensor out_159_axes_0 = const()[name = string("out_159_axes_0"), val = tensor([1])]; - fp16 var_6219_to_fp16 = const()[name = string("op_6219_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_159_cast_fp16 = layer_norm(axes = out_159_axes_0, epsilon = var_6219_to_fp16, x = inputs_159_cast_fp16)[name = string("out_159_cast_fp16")]; - tensor inputs_161_gamma_0_to_fp16 = const()[name = string("inputs_161_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210928320)))]; - tensor inputs_161_beta_0_to_fp16 = const()[name = string("inputs_161_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210930432)))]; - fp16 inputs_161_epsilon_0_to_fp16 = const()[name = string("inputs_161_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_161_cast_fp16 = batch_norm(beta = inputs_161_beta_0_to_fp16, epsilon = inputs_161_epsilon_0_to_fp16, gamma = inputs_161_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_159_cast_fp16)[name = string("inputs_161_cast_fp16")]; - int32 var_6233 = const()[name = string("op_6233"), val = int32(3)]; - tensor out_161_axes_0 = const()[name = string("out_161_axes_0"), val = tensor([1])]; - fp16 var_6264_to_fp16 = const()[name = string("op_6264_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_161_cast_fp16 = layer_norm(axes = out_161_axes_0, epsilon = var_6264_to_fp16, x = inputs_161_cast_fp16)[name = string("out_161_cast_fp16")]; - tensor input_433_gamma_0_to_fp16 = const()[name = string("input_433_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210932544)))]; - tensor input_433_beta_0_to_fp16 = const()[name = string("input_433_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210934656)))]; - fp16 input_433_epsilon_0_to_fp16 = const()[name = string("input_433_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_433_cast_fp16 = batch_norm(beta = input_433_beta_0_to_fp16, epsilon = input_433_epsilon_0_to_fp16, gamma = input_433_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_161_cast_fp16)[name = string("input_433_cast_fp16")]; - string var_6284_pad_type_0 = const()[name = string("op_6284_pad_type_0"), val = string("valid")]; - tensor var_6284_strides_0 = const()[name = string("op_6284_strides_0"), val = tensor([1, 1])]; - tensor var_6284_pad_0 = const()[name = string("op_6284_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6284_dilations_0 = const()[name = string("op_6284_dilations_0"), val = tensor([1, 1])]; - int32 var_6284_groups_0 = const()[name = string("op_6284_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210936768))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212509696))))[name = string("layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6284_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6284_dilations_0, groups = var_6284_groups_0, pad = var_6284_pad_0, pad_type = var_6284_pad_type_0, strides = var_6284_strides_0, weight = layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_433_cast_fp16)[name = string("op_6284_cast_fp16")]; - string var_6290_pad_type_0 = const()[name = string("op_6290_pad_type_0"), val = string("valid")]; - tensor var_6290_strides_0 = const()[name = string("op_6290_strides_0"), val = tensor([1, 1])]; - tensor var_6290_pad_0 = const()[name = string("op_6290_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6290_dilations_0 = const()[name = string("op_6290_dilations_0"), val = tensor([1, 1])]; - int32 var_6290_groups_0 = const()[name = string("op_6290_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212562880))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212513856))))[name = string("layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6290_cast_fp16 = conv(dilations = var_6290_dilations_0, groups = var_6290_groups_0, pad = var_6290_pad_0, pad_type = var_6290_pad_type_0, strides = var_6290_strides_0, weight = layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_433_cast_fp16)[name = string("op_6290_cast_fp16")]; - tensor input_435_cast_fp16 = add(x = var_6284_cast_fp16, y = var_6290_cast_fp16)[name = string("input_435_cast_fp16")]; - tensor input_437_cast_fp16 = silu(x = input_435_cast_fp16)[name = string("input_437_cast_fp16")]; - string var_6301_pad_type_0 = const()[name = string("op_6301_pad_type_0"), val = string("valid")]; - tensor var_6301_strides_0 = const()[name = string("op_6301_strides_0"), val = tensor([1, 1])]; - tensor var_6301_pad_0 = const()[name = string("op_6301_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6301_dilations_0 = const()[name = string("op_6301_dilations_0"), val = tensor([1, 1])]; - int32 var_6301_groups_0 = const()[name = string("op_6301_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213087232))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214660160))))[name = string("layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6301_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6301_dilations_0, groups = var_6301_groups_0, pad = var_6301_pad_0, pad_type = var_6301_pad_type_0, strides = var_6301_strides_0, weight = layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_437_cast_fp16)[name = string("op_6301_cast_fp16")]; - string var_6307_pad_type_0 = const()[name = string("op_6307_pad_type_0"), val = string("valid")]; - tensor var_6307_strides_0 = const()[name = string("op_6307_strides_0"), val = tensor([1, 1])]; - tensor var_6307_pad_0 = const()[name = string("op_6307_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6307_dilations_0 = const()[name = string("op_6307_dilations_0"), val = tensor([1, 1])]; - int32 var_6307_groups_0 = const()[name = string("op_6307_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214731520))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214661248))))[name = string("layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6307_cast_fp16 = conv(dilations = var_6307_dilations_0, groups = var_6307_groups_0, pad = var_6307_pad_0, pad_type = var_6307_pad_type_0, strides = var_6307_strides_0, weight = layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_437_cast_fp16)[name = string("op_6307_cast_fp16")]; - tensor x_99_cast_fp16 = add(x = var_6301_cast_fp16, y = var_6307_cast_fp16)[name = string("x_99_cast_fp16")]; - fp16 var_6309_to_fp16 = const()[name = string("op_6309_to_fp16"), val = fp16(0x1p-1)]; - tensor var_6310_cast_fp16 = mul(x = x_99_cast_fp16, y = var_6309_to_fp16)[name = string("op_6310_cast_fp16")]; - tensor inputs_163_cast_fp16 = add(x = inputs_161_cast_fp16, y = var_6310_cast_fp16)[name = string("inputs_163_cast_fp16")]; - tensor out_163_axes_0 = const()[name = string("out_163_axes_0"), val = tensor([1])]; - fp16 var_6320_to_fp16 = const()[name = string("op_6320_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_163_cast_fp16 = layer_norm(axes = out_163_axes_0, epsilon = var_6320_to_fp16, x = inputs_163_cast_fp16)[name = string("out_163_cast_fp16")]; - tensor obj_67_gamma_0_to_fp16 = const()[name = string("obj_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215255872)))]; - tensor obj_67_beta_0_to_fp16 = const()[name = string("obj_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215257984)))]; - fp16 obj_67_epsilon_0_to_fp16 = const()[name = string("obj_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_67_cast_fp16 = batch_norm(beta = obj_67_beta_0_to_fp16, epsilon = obj_67_epsilon_0_to_fp16, gamma = obj_67_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_163_cast_fp16)[name = string("obj_67_cast_fp16")]; - string var_6345_pad_type_0 = const()[name = string("op_6345_pad_type_0"), val = string("valid")]; - tensor var_6345_strides_0 = const()[name = string("op_6345_strides_0"), val = tensor([1, 1])]; - tensor var_6345_pad_0 = const()[name = string("op_6345_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6345_dilations_0 = const()[name = string("op_6345_dilations_0"), val = tensor([1, 1])]; - int32 var_6345_groups_0 = const()[name = string("op_6345_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215260096))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215653376))))[name = string("layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6345_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6345_dilations_0, groups = var_6345_groups_0, pad = var_6345_pad_0, pad_type = var_6345_pad_type_0, strides = var_6345_strides_0, weight = layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = string("op_6345_cast_fp16")]; - string var_6351_pad_type_0 = const()[name = string("op_6351_pad_type_0"), val = string("valid")]; - tensor var_6351_strides_0 = const()[name = string("op_6351_strides_0"), val = tensor([1, 1])]; - tensor var_6351_pad_0 = const()[name = string("op_6351_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6351_dilations_0 = const()[name = string("op_6351_dilations_0"), val = tensor([1, 1])]; - int32 var_6351_groups_0 = const()[name = string("op_6351_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215663872))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215654464))))[name = string("layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6351_cast_fp16 = conv(dilations = var_6351_dilations_0, groups = var_6351_groups_0, pad = var_6351_pad_0, pad_type = var_6351_pad_type_0, strides = var_6351_strides_0, weight = layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = string("op_6351_cast_fp16")]; - tensor query_65_cast_fp16 = add(x = var_6345_cast_fp16, y = var_6351_cast_fp16)[name = string("query_65_cast_fp16")]; - string var_6360_pad_type_0 = const()[name = string("op_6360_pad_type_0"), val = string("valid")]; - tensor var_6360_strides_0 = const()[name = string("op_6360_strides_0"), val = tensor([1, 1])]; - tensor var_6360_pad_0 = const()[name = string("op_6360_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6360_dilations_0 = const()[name = string("op_6360_dilations_0"), val = tensor([1, 1])]; - int32 var_6360_groups_0 = const()[name = string("op_6360_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215795008))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216188288))))[name = string("layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6360_cast_fp16 = conv(dilations = var_6360_dilations_0, groups = var_6360_groups_0, pad = var_6360_pad_0, pad_type = var_6360_pad_type_0, strides = var_6360_strides_0, weight = layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = string("op_6360_cast_fp16")]; - string var_6366_pad_type_0 = const()[name = string("op_6366_pad_type_0"), val = string("valid")]; - tensor var_6366_strides_0 = const()[name = string("op_6366_strides_0"), val = tensor([1, 1])]; - tensor var_6366_pad_0 = const()[name = string("op_6366_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6366_dilations_0 = const()[name = string("op_6366_dilations_0"), val = tensor([1, 1])]; - int32 var_6366_groups_0 = const()[name = string("op_6366_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216202688))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216189376))))[name = string("layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6366_cast_fp16 = conv(dilations = var_6366_dilations_0, groups = var_6366_groups_0, pad = var_6366_pad_0, pad_type = var_6366_pad_type_0, strides = var_6366_strides_0, weight = layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = string("op_6366_cast_fp16")]; - tensor key_33_cast_fp16 = add(x = var_6360_cast_fp16, y = var_6366_cast_fp16)[name = string("key_33_cast_fp16")]; - string var_6376_pad_type_0 = const()[name = string("op_6376_pad_type_0"), val = string("valid")]; - tensor var_6376_strides_0 = const()[name = string("op_6376_strides_0"), val = tensor([1, 1])]; - tensor var_6376_pad_0 = const()[name = string("op_6376_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6376_dilations_0 = const()[name = string("op_6376_dilations_0"), val = tensor([1, 1])]; - int32 var_6376_groups_0 = const()[name = string("op_6376_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216333824))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216727104))))[name = string("layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6376_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6376_dilations_0, groups = var_6376_groups_0, pad = var_6376_pad_0, pad_type = var_6376_pad_type_0, strides = var_6376_strides_0, weight = layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = string("op_6376_cast_fp16")]; - string var_6382_pad_type_0 = const()[name = string("op_6382_pad_type_0"), val = string("valid")]; - tensor var_6382_strides_0 = const()[name = string("op_6382_strides_0"), val = tensor([1, 1])]; - tensor var_6382_pad_0 = const()[name = string("op_6382_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6382_dilations_0 = const()[name = string("op_6382_dilations_0"), val = tensor([1, 1])]; - int32 var_6382_groups_0 = const()[name = string("op_6382_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216736000))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216728192))))[name = string("layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6382_cast_fp16 = conv(dilations = var_6382_dilations_0, groups = var_6382_groups_0, pad = var_6382_pad_0, pad_type = var_6382_pad_type_0, strides = var_6382_strides_0, weight = layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = string("op_6382_cast_fp16")]; - tensor value_33_cast_fp16 = add(x = var_6376_cast_fp16, y = var_6382_cast_fp16)[name = string("value_33_cast_fp16")]; - tensor var_6385_to_fp16 = const()[name = string("op_6385_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216867136)))]; - tensor query_67_cast_fp16 = add(x = query_65_cast_fp16, y = var_6385_to_fp16)[name = string("query_67_cast_fp16")]; - tensor var_6388_to_fp16 = const()[name = string("op_6388_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216869248)))]; - tensor q_with_bias_v_33_cast_fp16 = add(x = query_65_cast_fp16, y = var_6388_to_fp16)[name = string("q_with_bias_v_33_cast_fp16")]; - string var_6398_pad_type_0 = const()[name = string("op_6398_pad_type_0"), val = string("valid")]; - tensor var_6398_strides_0 = const()[name = string("op_6398_strides_0"), val = tensor([1, 1])]; - tensor var_6398_pad_0 = const()[name = string("op_6398_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6398_dilations_0 = const()[name = string("op_6398_dilations_0"), val = tensor([1, 1])]; - int32 var_6398_groups_0 = const()[name = string("op_6398_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216871360))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217264640))))[name = string("layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_6398_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6398_dilations_0, groups = var_6398_groups_0, pad = var_6398_pad_0, pad_type = var_6398_pad_type_0, strides = var_6398_strides_0, weight = layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_6398_cast_fp16")]; - string var_6404_pad_type_0 = const()[name = string("op_6404_pad_type_0"), val = string("valid")]; - tensor var_6404_strides_0 = const()[name = string("op_6404_strides_0"), val = tensor([1, 1])]; - tensor var_6404_pad_0 = const()[name = string("op_6404_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6404_dilations_0 = const()[name = string("op_6404_dilations_0"), val = tensor([1, 1])]; - int32 var_6404_groups_0 = const()[name = string("op_6404_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217300288))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217265728))))[name = string("layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6404_cast_fp16 = conv(dilations = var_6404_dilations_0, groups = var_6404_groups_0, pad = var_6404_pad_0, pad_type = var_6404_pad_type_0, strides = var_6404_strides_0, weight = layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_6404_cast_fp16")]; - tensor p_33_cast_fp16 = add(x = var_6398_cast_fp16, y = var_6404_cast_fp16)[name = string("p_33_cast_fp16")]; - tensor var_6408 = const()[name = string("op_6408"), val = tensor([1, 8, 128, 188])]; - tensor var_6409_cast_fp16 = reshape(shape = var_6408, x = q_with_bias_v_33_cast_fp16)[name = string("op_6409_cast_fp16")]; - tensor var_6410 = const()[name = string("op_6410"), val = tensor([1, 8, 128, -1])]; - tensor var_6411_cast_fp16 = reshape(shape = var_6410, x = p_33_cast_fp16)[name = string("op_6411_cast_fp16")]; - bool matrix_bd_129_transpose_x_0 = const()[name = string("matrix_bd_129_transpose_x_0"), val = bool(true)]; - bool matrix_bd_129_transpose_y_0 = const()[name = string("matrix_bd_129_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_129_cast_fp16 = matmul(transpose_x = matrix_bd_129_transpose_x_0, transpose_y = matrix_bd_129_transpose_y_0, x = var_6409_cast_fp16, y = var_6411_cast_fp16)[name = string("matrix_bd_129_cast_fp16")]; - tensor matrix_bd_131_pad_0 = const()[name = string("matrix_bd_131_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_131_mode_0 = const()[name = string("matrix_bd_131_mode_0"), val = string("constant")]; - fp16 const_186_to_fp16 = const()[name = string("const_186_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_131_cast_fp16 = pad(constant_val = const_186_to_fp16, mode = matrix_bd_131_mode_0, pad = matrix_bd_131_pad_0, x = matrix_bd_129_cast_fp16)[name = string("matrix_bd_131_cast_fp16")]; - tensor var_6420 = const()[name = string("op_6420"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_133_cast_fp16 = reshape(shape = var_6420, x = matrix_bd_131_cast_fp16)[name = string("matrix_bd_133_cast_fp16")]; - tensor var_6424_begin_0 = const()[name = string("op_6424_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_6424_end_0 = const()[name = string("op_6424_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_6424_end_mask_0 = const()[name = string("op_6424_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_6424_cast_fp16 = slice_by_index(begin = var_6424_begin_0, end = var_6424_end_0, end_mask = var_6424_end_mask_0, x = matrix_bd_133_cast_fp16)[name = string("op_6424_cast_fp16")]; - tensor var_6425 = const()[name = string("op_6425"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_135_cast_fp16 = reshape(shape = var_6425, x = var_6424_cast_fp16)[name = string("matrix_bd_135_cast_fp16")]; - tensor var_6430_begin_0 = const()[name = string("op_6430_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6430_end_0 = const()[name = string("op_6430_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_6430_end_mask_0 = const()[name = string("op_6430_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_6430_cast_fp16 = slice_by_index(begin = var_6430_begin_0, end = var_6430_end_0, end_mask = var_6430_end_mask_0, x = matrix_bd_135_cast_fp16)[name = string("op_6430_cast_fp16")]; - fp16 var_6431_to_fp16 = const()[name = string("op_6431_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_33_cast_fp16 = mul(x = var_6430_cast_fp16, y = var_6431_to_fp16)[name = string("qk_mask_33_cast_fp16")]; - tensor var_6435 = const()[name = string("op_6435"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_33_cast_fp16 = reshape(shape = var_6435, x = query_67_cast_fp16)[name = string("mh_q_33_cast_fp16")]; - fp16 var_6437_to_fp16 = const()[name = string("op_6437_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_6438_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_6437_to_fp16)[name = string("op_6438_cast_fp16")]; - tensor var_6441 = const()[name = string("op_6441"), val = tensor([1, 8, 128, 188])]; - tensor var_6442_cast_fp16 = reshape(shape = var_6441, x = key_33_cast_fp16)[name = string("op_6442_cast_fp16")]; - bool mh_w_65_transpose_x_0 = const()[name = string("mh_w_65_transpose_x_0"), val = bool(true)]; - bool mh_w_65_transpose_y_0 = const()[name = string("mh_w_65_transpose_y_0"), val = bool(false)]; - tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_6438_cast_fp16, y = var_6442_cast_fp16)[name = string("mh_w_65_cast_fp16")]; - tensor mh_w_67_cast_fp16 = add(x = mh_w_65_cast_fp16, y = qk_mask_33_cast_fp16)[name = string("mh_w_67_cast_fp16")]; - tensor var_6446_cast_fp16 = softmax(axis = var_6233, x = mh_w_67_cast_fp16)[name = string("op_6446_cast_fp16")]; - tensor var_6447 = const()[name = string("op_6447"), val = tensor([1, 8, 128, 188])]; - tensor var_6448_cast_fp16 = reshape(shape = var_6447, x = value_33_cast_fp16)[name = string("op_6448_cast_fp16")]; - bool attn_33_transpose_x_0 = const()[name = string("attn_33_transpose_x_0"), val = bool(false)]; - bool attn_33_transpose_y_0 = const()[name = string("attn_33_transpose_y_0"), val = bool(true)]; - tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_6448_cast_fp16, y = var_6446_cast_fp16)[name = string("attn_33_cast_fp16")]; - tensor var_6451 = const()[name = string("op_6451"), val = tensor([1, 1024, 1, 188])]; - tensor input_439_cast_fp16 = reshape(shape = var_6451, x = attn_33_cast_fp16)[name = string("input_439_cast_fp16")]; - string var_6461_pad_type_0 = const()[name = string("op_6461_pad_type_0"), val = string("valid")]; - tensor var_6461_strides_0 = const()[name = string("op_6461_strides_0"), val = tensor([1, 1])]; - tensor var_6461_pad_0 = const()[name = string("op_6461_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6461_dilations_0 = const()[name = string("op_6461_dilations_0"), val = tensor([1, 1])]; - int32 var_6461_groups_0 = const()[name = string("op_6461_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217431424))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217824704))))[name = string("layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6461_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6461_dilations_0, groups = var_6461_groups_0, pad = var_6461_pad_0, pad_type = var_6461_pad_type_0, strides = var_6461_strides_0, weight = layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_439_cast_fp16)[name = string("op_6461_cast_fp16")]; - string var_6467_pad_type_0 = const()[name = string("op_6467_pad_type_0"), val = string("valid")]; - tensor var_6467_strides_0 = const()[name = string("op_6467_strides_0"), val = tensor([1, 1])]; - tensor var_6467_pad_0 = const()[name = string("op_6467_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6467_dilations_0 = const()[name = string("op_6467_dilations_0"), val = tensor([1, 1])]; - int32 var_6467_groups_0 = const()[name = string("op_6467_groups_0"), val = int32(1)]; - tensor layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217834432))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217825792))))[name = string("layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6467_cast_fp16 = conv(dilations = var_6467_dilations_0, groups = var_6467_groups_0, pad = var_6467_pad_0, pad_type = var_6467_pad_type_0, strides = var_6467_strides_0, weight = layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_439_cast_fp16)[name = string("op_6467_cast_fp16")]; - tensor obj_69_cast_fp16 = add(x = var_6461_cast_fp16, y = var_6467_cast_fp16)[name = string("obj_69_cast_fp16")]; - tensor inputs_165_cast_fp16 = add(x = inputs_163_cast_fp16, y = obj_69_cast_fp16)[name = string("inputs_165_cast_fp16")]; - tensor out_165_axes_0 = const()[name = string("out_165_axes_0"), val = tensor([1])]; - fp16 var_6478_to_fp16 = const()[name = string("op_6478_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_165_cast_fp16 = layer_norm(axes = out_165_axes_0, epsilon = var_6478_to_fp16, x = inputs_165_cast_fp16)[name = string("out_165_cast_fp16")]; - tensor input_441_gamma_0_to_fp16 = const()[name = string("input_441_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217965568)))]; - tensor input_441_beta_0_to_fp16 = const()[name = string("input_441_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217967680)))]; - fp16 input_441_epsilon_0_to_fp16 = const()[name = string("input_441_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_441_cast_fp16 = batch_norm(beta = input_441_beta_0_to_fp16, epsilon = input_441_epsilon_0_to_fp16, gamma = input_441_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_165_cast_fp16)[name = string("input_441_cast_fp16")]; - string var_6499_pad_type_0 = const()[name = string("op_6499_pad_type_0"), val = string("valid")]; - tensor var_6499_strides_0 = const()[name = string("op_6499_strides_0"), val = tensor([1, 1])]; - tensor var_6499_pad_0 = const()[name = string("op_6499_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6499_dilations_0 = const()[name = string("op_6499_dilations_0"), val = tensor([1, 1])]; - int32 var_6499_groups_0 = const()[name = string("op_6499_groups_0"), val = int32(1)]; - tensor layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217969792))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218756288))))[name = string("layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6499_cast_fp16 = conv(dilations = var_6499_dilations_0, groups = var_6499_groups_0, pad = var_6499_pad_0, pad_type = var_6499_pad_type_0, strides = var_6499_strides_0, weight = layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_441_cast_fp16)[name = string("op_6499_cast_fp16")]; - string var_6505_pad_type_0 = const()[name = string("op_6505_pad_type_0"), val = string("valid")]; - tensor var_6505_strides_0 = const()[name = string("op_6505_strides_0"), val = tensor([1, 1])]; - tensor var_6505_pad_0 = const()[name = string("op_6505_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6505_dilations_0 = const()[name = string("op_6505_dilations_0"), val = tensor([1, 1])]; - int32 var_6505_groups_0 = const()[name = string("op_6505_groups_0"), val = int32(1)]; - tensor layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218778496))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218758400))))[name = string("layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6505_cast_fp16 = conv(dilations = var_6505_dilations_0, groups = var_6505_groups_0, pad = var_6505_pad_0, pad_type = var_6505_pad_type_0, strides = var_6505_strides_0, weight = layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_441_cast_fp16)[name = string("op_6505_cast_fp16")]; - tensor input_443_cast_fp16 = add(x = var_6499_cast_fp16, y = var_6505_cast_fp16)[name = string("input_443_cast_fp16")]; - int32 input_445_split_num_splits_0 = const()[name = string("input_445_split_num_splits_0"), val = int32(2)]; - int32 input_445_split_axis_0 = const()[name = string("input_445_split_axis_0"), val = int32(1)]; - tensor input_445_split_cast_fp16_0, tensor input_445_split_cast_fp16_1 = split(axis = input_445_split_axis_0, num_splits = input_445_split_num_splits_0, x = input_443_cast_fp16)[name = string("input_445_split_cast_fp16")]; - tensor input_445_split_1_sigmoid_cast_fp16 = sigmoid(x = input_445_split_cast_fp16_1)[name = string("input_445_split_1_sigmoid_cast_fp16")]; - tensor input_445_cast_fp16 = mul(x = input_445_split_cast_fp16_0, y = input_445_split_1_sigmoid_cast_fp16)[name = string("input_445_cast_fp16")]; - string input_447_pad_type_0 = const()[name = string("input_447_pad_type_0"), val = string("custom")]; - tensor input_447_pad_0 = const()[name = string("input_447_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_447_groups_0 = const()[name = string("input_447_groups_0"), val = int32(1024)]; - tensor input_447_strides_0 = const()[name = string("input_447_strides_0"), val = tensor([1, 1])]; - tensor input_447_dilations_0 = const()[name = string("input_447_dilations_0"), val = tensor([1, 1])]; - tensor const_300_to_fp16 = const()[name = string("const_300_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219040704)))]; - tensor const_301_to_fp16 = const()[name = string("const_301_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219059200)))]; - tensor input_449_cast_fp16 = conv(bias = const_301_to_fp16, dilations = input_447_dilations_0, groups = input_447_groups_0, pad = input_447_pad_0, pad_type = input_447_pad_type_0, strides = input_447_strides_0, weight = const_300_to_fp16, x = input_445_cast_fp16)[name = string("input_449_cast_fp16")]; - tensor input_451_cast_fp16 = silu(x = input_449_cast_fp16)[name = string("input_451_cast_fp16")]; - string var_6527_pad_type_0 = const()[name = string("op_6527_pad_type_0"), val = string("valid")]; - tensor var_6527_strides_0 = const()[name = string("op_6527_strides_0"), val = tensor([1, 1])]; - tensor var_6527_pad_0 = const()[name = string("op_6527_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6527_dilations_0 = const()[name = string("op_6527_dilations_0"), val = tensor([1, 1])]; - int32 var_6527_groups_0 = const()[name = string("op_6527_groups_0"), val = int32(1)]; - tensor layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219061312))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219454592))))[name = string("layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6527_cast_fp16 = conv(dilations = var_6527_dilations_0, groups = var_6527_groups_0, pad = var_6527_pad_0, pad_type = var_6527_pad_type_0, strides = var_6527_strides_0, weight = layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_451_cast_fp16)[name = string("op_6527_cast_fp16")]; - string var_6533_pad_type_0 = const()[name = string("op_6533_pad_type_0"), val = string("valid")]; - tensor var_6533_strides_0 = const()[name = string("op_6533_strides_0"), val = tensor([1, 1])]; - tensor var_6533_pad_0 = const()[name = string("op_6533_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6533_dilations_0 = const()[name = string("op_6533_dilations_0"), val = tensor([1, 1])]; - int32 var_6533_groups_0 = const()[name = string("op_6533_groups_0"), val = int32(1)]; - tensor layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219464256))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219455680))))[name = string("layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6533_cast_fp16 = conv(dilations = var_6533_dilations_0, groups = var_6533_groups_0, pad = var_6533_pad_0, pad_type = var_6533_pad_type_0, strides = var_6533_strides_0, weight = layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_451_cast_fp16)[name = string("op_6533_cast_fp16")]; - tensor x_101_cast_fp16 = add(x = var_6527_cast_fp16, y = var_6533_cast_fp16)[name = string("x_101_cast_fp16")]; - tensor inputs_167_cast_fp16 = add(x = inputs_165_cast_fp16, y = x_101_cast_fp16)[name = string("inputs_167_cast_fp16")]; - tensor out_167_axes_0 = const()[name = string("out_167_axes_0"), val = tensor([1])]; - fp16 var_6544_to_fp16 = const()[name = string("op_6544_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_167_cast_fp16 = layer_norm(axes = out_167_axes_0, epsilon = var_6544_to_fp16, x = inputs_167_cast_fp16)[name = string("out_167_cast_fp16")]; - tensor input_453_gamma_0_to_fp16 = const()[name = string("input_453_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219595392)))]; - tensor input_453_beta_0_to_fp16 = const()[name = string("input_453_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219597504)))]; - fp16 input_453_epsilon_0_to_fp16 = const()[name = string("input_453_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_453_cast_fp16 = batch_norm(beta = input_453_beta_0_to_fp16, epsilon = input_453_epsilon_0_to_fp16, gamma = input_453_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_167_cast_fp16)[name = string("input_453_cast_fp16")]; - string var_6564_pad_type_0 = const()[name = string("op_6564_pad_type_0"), val = string("valid")]; - tensor var_6564_strides_0 = const()[name = string("op_6564_strides_0"), val = tensor([1, 1])]; - tensor var_6564_pad_0 = const()[name = string("op_6564_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6564_dilations_0 = const()[name = string("op_6564_dilations_0"), val = tensor([1, 1])]; - int32 var_6564_groups_0 = const()[name = string("op_6564_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219599616))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221172544))))[name = string("layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6564_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6564_dilations_0, groups = var_6564_groups_0, pad = var_6564_pad_0, pad_type = var_6564_pad_type_0, strides = var_6564_strides_0, weight = layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_453_cast_fp16)[name = string("op_6564_cast_fp16")]; - string var_6570_pad_type_0 = const()[name = string("op_6570_pad_type_0"), val = string("valid")]; - tensor var_6570_strides_0 = const()[name = string("op_6570_strides_0"), val = tensor([1, 1])]; - tensor var_6570_pad_0 = const()[name = string("op_6570_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6570_dilations_0 = const()[name = string("op_6570_dilations_0"), val = tensor([1, 1])]; - int32 var_6570_groups_0 = const()[name = string("op_6570_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221216128))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221176704))))[name = string("layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6570_cast_fp16 = conv(dilations = var_6570_dilations_0, groups = var_6570_groups_0, pad = var_6570_pad_0, pad_type = var_6570_pad_type_0, strides = var_6570_strides_0, weight = layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_453_cast_fp16)[name = string("op_6570_cast_fp16")]; - tensor input_455_cast_fp16 = add(x = var_6564_cast_fp16, y = var_6570_cast_fp16)[name = string("input_455_cast_fp16")]; - tensor input_457_cast_fp16 = silu(x = input_455_cast_fp16)[name = string("input_457_cast_fp16")]; - string var_6581_pad_type_0 = const()[name = string("op_6581_pad_type_0"), val = string("valid")]; - tensor var_6581_strides_0 = const()[name = string("op_6581_strides_0"), val = tensor([1, 1])]; - tensor var_6581_pad_0 = const()[name = string("op_6581_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6581_dilations_0 = const()[name = string("op_6581_dilations_0"), val = tensor([1, 1])]; - int32 var_6581_groups_0 = const()[name = string("op_6581_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221740480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223313408))))[name = string("layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6581_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6581_dilations_0, groups = var_6581_groups_0, pad = var_6581_pad_0, pad_type = var_6581_pad_type_0, strides = var_6581_strides_0, weight = layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_457_cast_fp16)[name = string("op_6581_cast_fp16")]; - string var_6587_pad_type_0 = const()[name = string("op_6587_pad_type_0"), val = string("valid")]; - tensor var_6587_strides_0 = const()[name = string("op_6587_strides_0"), val = tensor([1, 1])]; - tensor var_6587_pad_0 = const()[name = string("op_6587_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6587_dilations_0 = const()[name = string("op_6587_dilations_0"), val = tensor([1, 1])]; - int32 var_6587_groups_0 = const()[name = string("op_6587_groups_0"), val = int32(1)]; - tensor layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223359296))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223314496))))[name = string("layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6587_cast_fp16 = conv(dilations = var_6587_dilations_0, groups = var_6587_groups_0, pad = var_6587_pad_0, pad_type = var_6587_pad_type_0, strides = var_6587_strides_0, weight = layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_457_cast_fp16)[name = string("op_6587_cast_fp16")]; - tensor x_103_cast_fp16 = add(x = var_6581_cast_fp16, y = var_6587_cast_fp16)[name = string("x_103_cast_fp16")]; - fp16 var_6589_to_fp16 = const()[name = string("op_6589_to_fp16"), val = fp16(0x1p-1)]; - tensor var_6590_cast_fp16 = mul(x = x_103_cast_fp16, y = var_6589_to_fp16)[name = string("op_6590_cast_fp16")]; - tensor inputs_169_cast_fp16 = add(x = inputs_167_cast_fp16, y = var_6590_cast_fp16)[name = string("inputs_169_cast_fp16")]; - tensor out_169_axes_0 = const()[name = string("out_169_axes_0"), val = tensor([1])]; - fp16 var_6600_to_fp16 = const()[name = string("op_6600_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_169_cast_fp16 = layer_norm(axes = out_169_axes_0, epsilon = var_6600_to_fp16, x = inputs_169_cast_fp16)[name = string("out_169_cast_fp16")]; - tensor inputs_171_gamma_0_to_fp16 = const()[name = string("inputs_171_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223883648)))]; - tensor inputs_171_beta_0_to_fp16 = const()[name = string("inputs_171_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223885760)))]; - fp16 inputs_171_epsilon_0_to_fp16 = const()[name = string("inputs_171_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_171_cast_fp16 = batch_norm(beta = inputs_171_beta_0_to_fp16, epsilon = inputs_171_epsilon_0_to_fp16, gamma = inputs_171_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_169_cast_fp16)[name = string("inputs_171_cast_fp16")]; - int32 var_6614 = const()[name = string("op_6614"), val = int32(3)]; - tensor out_171_axes_0 = const()[name = string("out_171_axes_0"), val = tensor([1])]; - fp16 var_6645_to_fp16 = const()[name = string("op_6645_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_171_cast_fp16 = layer_norm(axes = out_171_axes_0, epsilon = var_6645_to_fp16, x = inputs_171_cast_fp16)[name = string("out_171_cast_fp16")]; - tensor input_459_gamma_0_to_fp16 = const()[name = string("input_459_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223887872)))]; - tensor input_459_beta_0_to_fp16 = const()[name = string("input_459_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223889984)))]; - fp16 input_459_epsilon_0_to_fp16 = const()[name = string("input_459_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_459_cast_fp16 = batch_norm(beta = input_459_beta_0_to_fp16, epsilon = input_459_epsilon_0_to_fp16, gamma = input_459_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_171_cast_fp16)[name = string("input_459_cast_fp16")]; - string var_6665_pad_type_0 = const()[name = string("op_6665_pad_type_0"), val = string("valid")]; - tensor var_6665_strides_0 = const()[name = string("op_6665_strides_0"), val = tensor([1, 1])]; - tensor var_6665_pad_0 = const()[name = string("op_6665_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6665_dilations_0 = const()[name = string("op_6665_dilations_0"), val = tensor([1, 1])]; - int32 var_6665_groups_0 = const()[name = string("op_6665_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223892096))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225465024))))[name = string("layers_17_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6665_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6665_dilations_0, groups = var_6665_groups_0, pad = var_6665_pad_0, pad_type = var_6665_pad_type_0, strides = var_6665_strides_0, weight = layers_17_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_459_cast_fp16)[name = string("op_6665_cast_fp16")]; - string var_6671_pad_type_0 = const()[name = string("op_6671_pad_type_0"), val = string("valid")]; - tensor var_6671_strides_0 = const()[name = string("op_6671_strides_0"), val = tensor([1, 1])]; - tensor var_6671_pad_0 = const()[name = string("op_6671_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6671_dilations_0 = const()[name = string("op_6671_dilations_0"), val = tensor([1, 1])]; - int32 var_6671_groups_0 = const()[name = string("op_6671_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225513216))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225469184))))[name = string("layers_17_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6671_cast_fp16 = conv(dilations = var_6671_dilations_0, groups = var_6671_groups_0, pad = var_6671_pad_0, pad_type = var_6671_pad_type_0, strides = var_6671_strides_0, weight = layers_17_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_459_cast_fp16)[name = string("op_6671_cast_fp16")]; - tensor input_461_cast_fp16 = add(x = var_6665_cast_fp16, y = var_6671_cast_fp16)[name = string("input_461_cast_fp16")]; - tensor input_463_cast_fp16 = silu(x = input_461_cast_fp16)[name = string("input_463_cast_fp16")]; - string var_6682_pad_type_0 = const()[name = string("op_6682_pad_type_0"), val = string("valid")]; - tensor var_6682_strides_0 = const()[name = string("op_6682_strides_0"), val = tensor([1, 1])]; - tensor var_6682_pad_0 = const()[name = string("op_6682_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6682_dilations_0 = const()[name = string("op_6682_dilations_0"), val = tensor([1, 1])]; - int32 var_6682_groups_0 = const()[name = string("op_6682_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226037568))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227610496))))[name = string("layers_17_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6682_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6682_dilations_0, groups = var_6682_groups_0, pad = var_6682_pad_0, pad_type = var_6682_pad_type_0, strides = var_6682_strides_0, weight = layers_17_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_463_cast_fp16)[name = string("op_6682_cast_fp16")]; - string var_6688_pad_type_0 = const()[name = string("op_6688_pad_type_0"), val = string("valid")]; - tensor var_6688_strides_0 = const()[name = string("op_6688_strides_0"), val = tensor([1, 1])]; - tensor var_6688_pad_0 = const()[name = string("op_6688_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6688_dilations_0 = const()[name = string("op_6688_dilations_0"), val = tensor([1, 1])]; - int32 var_6688_groups_0 = const()[name = string("op_6688_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227660096))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227611584))))[name = string("layers_17_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6688_cast_fp16 = conv(dilations = var_6688_dilations_0, groups = var_6688_groups_0, pad = var_6688_pad_0, pad_type = var_6688_pad_type_0, strides = var_6688_strides_0, weight = layers_17_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_463_cast_fp16)[name = string("op_6688_cast_fp16")]; - tensor x_105_cast_fp16 = add(x = var_6682_cast_fp16, y = var_6688_cast_fp16)[name = string("x_105_cast_fp16")]; - fp16 var_6690_to_fp16 = const()[name = string("op_6690_to_fp16"), val = fp16(0x1p-1)]; - tensor var_6691_cast_fp16 = mul(x = x_105_cast_fp16, y = var_6690_to_fp16)[name = string("op_6691_cast_fp16")]; - tensor inputs_173_cast_fp16 = add(x = inputs_171_cast_fp16, y = var_6691_cast_fp16)[name = string("inputs_173_cast_fp16")]; - tensor out_173_axes_0 = const()[name = string("out_173_axes_0"), val = tensor([1])]; - fp16 var_6701_to_fp16 = const()[name = string("op_6701_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_173_cast_fp16 = layer_norm(axes = out_173_axes_0, epsilon = var_6701_to_fp16, x = inputs_173_cast_fp16)[name = string("out_173_cast_fp16")]; - tensor obj_71_gamma_0_to_fp16 = const()[name = string("obj_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228184448)))]; - tensor obj_71_beta_0_to_fp16 = const()[name = string("obj_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228186560)))]; - fp16 obj_71_epsilon_0_to_fp16 = const()[name = string("obj_71_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_71_cast_fp16 = batch_norm(beta = obj_71_beta_0_to_fp16, epsilon = obj_71_epsilon_0_to_fp16, gamma = obj_71_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_173_cast_fp16)[name = string("obj_71_cast_fp16")]; - string var_6726_pad_type_0 = const()[name = string("op_6726_pad_type_0"), val = string("valid")]; - tensor var_6726_strides_0 = const()[name = string("op_6726_strides_0"), val = tensor([1, 1])]; - tensor var_6726_pad_0 = const()[name = string("op_6726_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6726_dilations_0 = const()[name = string("op_6726_dilations_0"), val = tensor([1, 1])]; - int32 var_6726_groups_0 = const()[name = string("op_6726_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228188672))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228581952))))[name = string("layers_17_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6726_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6726_dilations_0, groups = var_6726_groups_0, pad = var_6726_pad_0, pad_type = var_6726_pad_type_0, strides = var_6726_strides_0, weight = layers_17_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = string("op_6726_cast_fp16")]; - string var_6732_pad_type_0 = const()[name = string("op_6732_pad_type_0"), val = string("valid")]; - tensor var_6732_strides_0 = const()[name = string("op_6732_strides_0"), val = tensor([1, 1])]; - tensor var_6732_pad_0 = const()[name = string("op_6732_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6732_dilations_0 = const()[name = string("op_6732_dilations_0"), val = tensor([1, 1])]; - int32 var_6732_groups_0 = const()[name = string("op_6732_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228594624))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228583040))))[name = string("layers_17_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6732_cast_fp16 = conv(dilations = var_6732_dilations_0, groups = var_6732_groups_0, pad = var_6732_pad_0, pad_type = var_6732_pad_type_0, strides = var_6732_strides_0, weight = layers_17_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_71_cast_fp16)[name = string("op_6732_cast_fp16")]; - tensor query_69_cast_fp16 = add(x = var_6726_cast_fp16, y = var_6732_cast_fp16)[name = string("query_69_cast_fp16")]; - string var_6741_pad_type_0 = const()[name = string("op_6741_pad_type_0"), val = string("valid")]; - tensor var_6741_strides_0 = const()[name = string("op_6741_strides_0"), val = tensor([1, 1])]; - tensor var_6741_pad_0 = const()[name = string("op_6741_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6741_dilations_0 = const()[name = string("op_6741_dilations_0"), val = tensor([1, 1])]; - int32 var_6741_groups_0 = const()[name = string("op_6741_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228725760))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229119040))))[name = string("layers_17_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6741_cast_fp16 = conv(dilations = var_6741_dilations_0, groups = var_6741_groups_0, pad = var_6741_pad_0, pad_type = var_6741_pad_type_0, strides = var_6741_strides_0, weight = layers_17_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = string("op_6741_cast_fp16")]; - string var_6747_pad_type_0 = const()[name = string("op_6747_pad_type_0"), val = string("valid")]; - tensor var_6747_strides_0 = const()[name = string("op_6747_strides_0"), val = tensor([1, 1])]; - tensor var_6747_pad_0 = const()[name = string("op_6747_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6747_dilations_0 = const()[name = string("op_6747_dilations_0"), val = tensor([1, 1])]; - int32 var_6747_groups_0 = const()[name = string("op_6747_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229134144))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229120128))))[name = string("layers_17_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6747_cast_fp16 = conv(dilations = var_6747_dilations_0, groups = var_6747_groups_0, pad = var_6747_pad_0, pad_type = var_6747_pad_type_0, strides = var_6747_strides_0, weight = layers_17_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_71_cast_fp16)[name = string("op_6747_cast_fp16")]; - tensor key_35_cast_fp16 = add(x = var_6741_cast_fp16, y = var_6747_cast_fp16)[name = string("key_35_cast_fp16")]; - string var_6757_pad_type_0 = const()[name = string("op_6757_pad_type_0"), val = string("valid")]; - tensor var_6757_strides_0 = const()[name = string("op_6757_strides_0"), val = tensor([1, 1])]; - tensor var_6757_pad_0 = const()[name = string("op_6757_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6757_dilations_0 = const()[name = string("op_6757_dilations_0"), val = tensor([1, 1])]; - int32 var_6757_groups_0 = const()[name = string("op_6757_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229265280))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229658560))))[name = string("layers_17_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6757_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6757_dilations_0, groups = var_6757_groups_0, pad = var_6757_pad_0, pad_type = var_6757_pad_type_0, strides = var_6757_strides_0, weight = layers_17_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = string("op_6757_cast_fp16")]; - string var_6763_pad_type_0 = const()[name = string("op_6763_pad_type_0"), val = string("valid")]; - tensor var_6763_strides_0 = const()[name = string("op_6763_strides_0"), val = tensor([1, 1])]; - tensor var_6763_pad_0 = const()[name = string("op_6763_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6763_dilations_0 = const()[name = string("op_6763_dilations_0"), val = tensor([1, 1])]; - int32 var_6763_groups_0 = const()[name = string("op_6763_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229668928))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229659648))))[name = string("layers_17_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6763_cast_fp16 = conv(dilations = var_6763_dilations_0, groups = var_6763_groups_0, pad = var_6763_pad_0, pad_type = var_6763_pad_type_0, strides = var_6763_strides_0, weight = layers_17_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_71_cast_fp16)[name = string("op_6763_cast_fp16")]; - tensor value_35_cast_fp16 = add(x = var_6757_cast_fp16, y = var_6763_cast_fp16)[name = string("value_35_cast_fp16")]; - tensor var_6766_to_fp16 = const()[name = string("op_6766_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229800064)))]; - tensor query_71_cast_fp16 = add(x = query_69_cast_fp16, y = var_6766_to_fp16)[name = string("query_71_cast_fp16")]; - tensor var_6769_to_fp16 = const()[name = string("op_6769_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229802176)))]; - tensor q_with_bias_v_35_cast_fp16 = add(x = query_69_cast_fp16, y = var_6769_to_fp16)[name = string("q_with_bias_v_35_cast_fp16")]; - string var_6779_pad_type_0 = const()[name = string("op_6779_pad_type_0"), val = string("valid")]; - tensor var_6779_strides_0 = const()[name = string("op_6779_strides_0"), val = tensor([1, 1])]; - tensor var_6779_pad_0 = const()[name = string("op_6779_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6779_dilations_0 = const()[name = string("op_6779_dilations_0"), val = tensor([1, 1])]; - int32 var_6779_groups_0 = const()[name = string("op_6779_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229804288))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230197568))))[name = string("layers_17_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_6779_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6779_dilations_0, groups = var_6779_groups_0, pad = var_6779_pad_0, pad_type = var_6779_pad_type_0, strides = var_6779_strides_0, weight = layers_17_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_6779_cast_fp16")]; - string var_6785_pad_type_0 = const()[name = string("op_6785_pad_type_0"), val = string("valid")]; - tensor var_6785_strides_0 = const()[name = string("op_6785_strides_0"), val = tensor([1, 1])]; - tensor var_6785_pad_0 = const()[name = string("op_6785_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6785_dilations_0 = const()[name = string("op_6785_dilations_0"), val = tensor([1, 1])]; - int32 var_6785_groups_0 = const()[name = string("op_6785_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230227776))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230198656))))[name = string("layers_17_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6785_cast_fp16 = conv(dilations = var_6785_dilations_0, groups = var_6785_groups_0, pad = var_6785_pad_0, pad_type = var_6785_pad_type_0, strides = var_6785_strides_0, weight = layers_17_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_6785_cast_fp16")]; - tensor p_35_cast_fp16 = add(x = var_6779_cast_fp16, y = var_6785_cast_fp16)[name = string("p_35_cast_fp16")]; - tensor var_6789 = const()[name = string("op_6789"), val = tensor([1, 8, 128, 188])]; - tensor var_6790_cast_fp16 = reshape(shape = var_6789, x = q_with_bias_v_35_cast_fp16)[name = string("op_6790_cast_fp16")]; - tensor var_6791 = const()[name = string("op_6791"), val = tensor([1, 8, 128, -1])]; - tensor var_6792_cast_fp16 = reshape(shape = var_6791, x = p_35_cast_fp16)[name = string("op_6792_cast_fp16")]; - bool matrix_bd_137_transpose_x_0 = const()[name = string("matrix_bd_137_transpose_x_0"), val = bool(true)]; - bool matrix_bd_137_transpose_y_0 = const()[name = string("matrix_bd_137_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_137_cast_fp16 = matmul(transpose_x = matrix_bd_137_transpose_x_0, transpose_y = matrix_bd_137_transpose_y_0, x = var_6790_cast_fp16, y = var_6792_cast_fp16)[name = string("matrix_bd_137_cast_fp16")]; - tensor matrix_bd_139_pad_0 = const()[name = string("matrix_bd_139_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_139_mode_0 = const()[name = string("matrix_bd_139_mode_0"), val = string("constant")]; - fp16 const_197_to_fp16 = const()[name = string("const_197_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_139_cast_fp16 = pad(constant_val = const_197_to_fp16, mode = matrix_bd_139_mode_0, pad = matrix_bd_139_pad_0, x = matrix_bd_137_cast_fp16)[name = string("matrix_bd_139_cast_fp16")]; - tensor var_6801 = const()[name = string("op_6801"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_141_cast_fp16 = reshape(shape = var_6801, x = matrix_bd_139_cast_fp16)[name = string("matrix_bd_141_cast_fp16")]; - tensor var_6805_begin_0 = const()[name = string("op_6805_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_6805_end_0 = const()[name = string("op_6805_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_6805_end_mask_0 = const()[name = string("op_6805_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_6805_cast_fp16 = slice_by_index(begin = var_6805_begin_0, end = var_6805_end_0, end_mask = var_6805_end_mask_0, x = matrix_bd_141_cast_fp16)[name = string("op_6805_cast_fp16")]; - tensor var_6806 = const()[name = string("op_6806"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_143_cast_fp16 = reshape(shape = var_6806, x = var_6805_cast_fp16)[name = string("matrix_bd_143_cast_fp16")]; - tensor var_6811_begin_0 = const()[name = string("op_6811_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6811_end_0 = const()[name = string("op_6811_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_6811_end_mask_0 = const()[name = string("op_6811_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_6811_cast_fp16 = slice_by_index(begin = var_6811_begin_0, end = var_6811_end_0, end_mask = var_6811_end_mask_0, x = matrix_bd_143_cast_fp16)[name = string("op_6811_cast_fp16")]; - fp16 var_6812_to_fp16 = const()[name = string("op_6812_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_35_cast_fp16 = mul(x = var_6811_cast_fp16, y = var_6812_to_fp16)[name = string("qk_mask_35_cast_fp16")]; - tensor var_6816 = const()[name = string("op_6816"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_35_cast_fp16 = reshape(shape = var_6816, x = query_71_cast_fp16)[name = string("mh_q_35_cast_fp16")]; - fp16 var_6818_to_fp16 = const()[name = string("op_6818_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_6819_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_6818_to_fp16)[name = string("op_6819_cast_fp16")]; - tensor var_6822 = const()[name = string("op_6822"), val = tensor([1, 8, 128, 188])]; - tensor var_6823_cast_fp16 = reshape(shape = var_6822, x = key_35_cast_fp16)[name = string("op_6823_cast_fp16")]; - bool mh_w_69_transpose_x_0 = const()[name = string("mh_w_69_transpose_x_0"), val = bool(true)]; - bool mh_w_69_transpose_y_0 = const()[name = string("mh_w_69_transpose_y_0"), val = bool(false)]; - tensor mh_w_69_cast_fp16 = matmul(transpose_x = mh_w_69_transpose_x_0, transpose_y = mh_w_69_transpose_y_0, x = var_6819_cast_fp16, y = var_6823_cast_fp16)[name = string("mh_w_69_cast_fp16")]; - tensor mh_w_71_cast_fp16 = add(x = mh_w_69_cast_fp16, y = qk_mask_35_cast_fp16)[name = string("mh_w_71_cast_fp16")]; - tensor var_6827_cast_fp16 = softmax(axis = var_6614, x = mh_w_71_cast_fp16)[name = string("op_6827_cast_fp16")]; - tensor var_6828 = const()[name = string("op_6828"), val = tensor([1, 8, 128, 188])]; - tensor var_6829_cast_fp16 = reshape(shape = var_6828, x = value_35_cast_fp16)[name = string("op_6829_cast_fp16")]; - bool attn_35_transpose_x_0 = const()[name = string("attn_35_transpose_x_0"), val = bool(false)]; - bool attn_35_transpose_y_0 = const()[name = string("attn_35_transpose_y_0"), val = bool(true)]; - tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_6829_cast_fp16, y = var_6827_cast_fp16)[name = string("attn_35_cast_fp16")]; - tensor var_6832 = const()[name = string("op_6832"), val = tensor([1, 1024, 1, 188])]; - tensor input_465_cast_fp16 = reshape(shape = var_6832, x = attn_35_cast_fp16)[name = string("input_465_cast_fp16")]; - string var_6842_pad_type_0 = const()[name = string("op_6842_pad_type_0"), val = string("valid")]; - tensor var_6842_strides_0 = const()[name = string("op_6842_strides_0"), val = tensor([1, 1])]; - tensor var_6842_pad_0 = const()[name = string("op_6842_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6842_dilations_0 = const()[name = string("op_6842_dilations_0"), val = tensor([1, 1])]; - int32 var_6842_groups_0 = const()[name = string("op_6842_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230358912))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230752192))))[name = string("layers_17_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_6842_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6842_dilations_0, groups = var_6842_groups_0, pad = var_6842_pad_0, pad_type = var_6842_pad_type_0, strides = var_6842_strides_0, weight = layers_17_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_465_cast_fp16)[name = string("op_6842_cast_fp16")]; - string var_6848_pad_type_0 = const()[name = string("op_6848_pad_type_0"), val = string("valid")]; - tensor var_6848_strides_0 = const()[name = string("op_6848_strides_0"), val = tensor([1, 1])]; - tensor var_6848_pad_0 = const()[name = string("op_6848_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6848_dilations_0 = const()[name = string("op_6848_dilations_0"), val = tensor([1, 1])]; - int32 var_6848_groups_0 = const()[name = string("op_6848_groups_0"), val = int32(1)]; - tensor layers_17_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230761408))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230753280))))[name = string("layers_17_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6848_cast_fp16 = conv(dilations = var_6848_dilations_0, groups = var_6848_groups_0, pad = var_6848_pad_0, pad_type = var_6848_pad_type_0, strides = var_6848_strides_0, weight = layers_17_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_465_cast_fp16)[name = string("op_6848_cast_fp16")]; - tensor obj_73_cast_fp16 = add(x = var_6842_cast_fp16, y = var_6848_cast_fp16)[name = string("obj_73_cast_fp16")]; - tensor inputs_175_cast_fp16 = add(x = inputs_173_cast_fp16, y = obj_73_cast_fp16)[name = string("inputs_175_cast_fp16")]; - tensor out_175_axes_0 = const()[name = string("out_175_axes_0"), val = tensor([1])]; - fp16 var_6859_to_fp16 = const()[name = string("op_6859_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_175_cast_fp16 = layer_norm(axes = out_175_axes_0, epsilon = var_6859_to_fp16, x = inputs_175_cast_fp16)[name = string("out_175_cast_fp16")]; - tensor input_467_gamma_0_to_fp16 = const()[name = string("input_467_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230892544)))]; - tensor input_467_beta_0_to_fp16 = const()[name = string("input_467_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230894656)))]; - fp16 input_467_epsilon_0_to_fp16 = const()[name = string("input_467_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_467_cast_fp16 = batch_norm(beta = input_467_beta_0_to_fp16, epsilon = input_467_epsilon_0_to_fp16, gamma = input_467_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_175_cast_fp16)[name = string("input_467_cast_fp16")]; - string var_6880_pad_type_0 = const()[name = string("op_6880_pad_type_0"), val = string("valid")]; - tensor var_6880_strides_0 = const()[name = string("op_6880_strides_0"), val = tensor([1, 1])]; - tensor var_6880_pad_0 = const()[name = string("op_6880_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6880_dilations_0 = const()[name = string("op_6880_dilations_0"), val = tensor([1, 1])]; - int32 var_6880_groups_0 = const()[name = string("op_6880_groups_0"), val = int32(1)]; - tensor layers_17_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230896768))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231683264))))[name = string("layers_17_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6880_cast_fp16 = conv(dilations = var_6880_dilations_0, groups = var_6880_groups_0, pad = var_6880_pad_0, pad_type = var_6880_pad_type_0, strides = var_6880_strides_0, weight = layers_17_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_467_cast_fp16)[name = string("op_6880_cast_fp16")]; - string var_6886_pad_type_0 = const()[name = string("op_6886_pad_type_0"), val = string("valid")]; - tensor var_6886_strides_0 = const()[name = string("op_6886_strides_0"), val = tensor([1, 1])]; - tensor var_6886_pad_0 = const()[name = string("op_6886_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6886_dilations_0 = const()[name = string("op_6886_dilations_0"), val = tensor([1, 1])]; - int32 var_6886_groups_0 = const()[name = string("op_6886_groups_0"), val = int32(1)]; - tensor layers_17_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231706560))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231685376))))[name = string("layers_17_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6886_cast_fp16 = conv(dilations = var_6886_dilations_0, groups = var_6886_groups_0, pad = var_6886_pad_0, pad_type = var_6886_pad_type_0, strides = var_6886_strides_0, weight = layers_17_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_467_cast_fp16)[name = string("op_6886_cast_fp16")]; - tensor input_469_cast_fp16 = add(x = var_6880_cast_fp16, y = var_6886_cast_fp16)[name = string("input_469_cast_fp16")]; - int32 input_471_split_num_splits_0 = const()[name = string("input_471_split_num_splits_0"), val = int32(2)]; - int32 input_471_split_axis_0 = const()[name = string("input_471_split_axis_0"), val = int32(1)]; - tensor input_471_split_cast_fp16_0, tensor input_471_split_cast_fp16_1 = split(axis = input_471_split_axis_0, num_splits = input_471_split_num_splits_0, x = input_469_cast_fp16)[name = string("input_471_split_cast_fp16")]; - tensor input_471_split_1_sigmoid_cast_fp16 = sigmoid(x = input_471_split_cast_fp16_1)[name = string("input_471_split_1_sigmoid_cast_fp16")]; - tensor input_471_cast_fp16 = mul(x = input_471_split_cast_fp16_0, y = input_471_split_1_sigmoid_cast_fp16)[name = string("input_471_cast_fp16")]; - string input_473_pad_type_0 = const()[name = string("input_473_pad_type_0"), val = string("custom")]; - tensor input_473_pad_0 = const()[name = string("input_473_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_473_groups_0 = const()[name = string("input_473_groups_0"), val = int32(1024)]; - tensor input_473_strides_0 = const()[name = string("input_473_strides_0"), val = tensor([1, 1])]; - tensor input_473_dilations_0 = const()[name = string("input_473_dilations_0"), val = tensor([1, 1])]; - tensor const_302_to_fp16 = const()[name = string("const_302_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231968768)))]; - tensor const_303_to_fp16 = const()[name = string("const_303_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231987264)))]; - tensor input_475_cast_fp16 = conv(bias = const_303_to_fp16, dilations = input_473_dilations_0, groups = input_473_groups_0, pad = input_473_pad_0, pad_type = input_473_pad_type_0, strides = input_473_strides_0, weight = const_302_to_fp16, x = input_471_cast_fp16)[name = string("input_475_cast_fp16")]; - tensor input_477_cast_fp16 = silu(x = input_475_cast_fp16)[name = string("input_477_cast_fp16")]; - string var_6908_pad_type_0 = const()[name = string("op_6908_pad_type_0"), val = string("valid")]; - tensor var_6908_strides_0 = const()[name = string("op_6908_strides_0"), val = tensor([1, 1])]; - tensor var_6908_pad_0 = const()[name = string("op_6908_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6908_dilations_0 = const()[name = string("op_6908_dilations_0"), val = tensor([1, 1])]; - int32 var_6908_groups_0 = const()[name = string("op_6908_groups_0"), val = int32(1)]; - tensor layers_17_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231989376))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232382656))))[name = string("layers_17_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6908_cast_fp16 = conv(dilations = var_6908_dilations_0, groups = var_6908_groups_0, pad = var_6908_pad_0, pad_type = var_6908_pad_type_0, strides = var_6908_strides_0, weight = layers_17_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_477_cast_fp16)[name = string("op_6908_cast_fp16")]; - string var_6914_pad_type_0 = const()[name = string("op_6914_pad_type_0"), val = string("valid")]; - tensor var_6914_strides_0 = const()[name = string("op_6914_strides_0"), val = tensor([1, 1])]; - tensor var_6914_pad_0 = const()[name = string("op_6914_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6914_dilations_0 = const()[name = string("op_6914_dilations_0"), val = tensor([1, 1])]; - int32 var_6914_groups_0 = const()[name = string("op_6914_groups_0"), val = int32(1)]; - tensor layers_17_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232392640))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232383744))))[name = string("layers_17_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6914_cast_fp16 = conv(dilations = var_6914_dilations_0, groups = var_6914_groups_0, pad = var_6914_pad_0, pad_type = var_6914_pad_type_0, strides = var_6914_strides_0, weight = layers_17_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_477_cast_fp16)[name = string("op_6914_cast_fp16")]; - tensor x_107_cast_fp16 = add(x = var_6908_cast_fp16, y = var_6914_cast_fp16)[name = string("x_107_cast_fp16")]; - tensor inputs_177_cast_fp16 = add(x = inputs_175_cast_fp16, y = x_107_cast_fp16)[name = string("inputs_177_cast_fp16")]; - tensor out_177_axes_0 = const()[name = string("out_177_axes_0"), val = tensor([1])]; - fp16 var_6925_to_fp16 = const()[name = string("op_6925_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_177_cast_fp16 = layer_norm(axes = out_177_axes_0, epsilon = var_6925_to_fp16, x = inputs_177_cast_fp16)[name = string("out_177_cast_fp16")]; - tensor input_479_gamma_0_to_fp16 = const()[name = string("input_479_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232523776)))]; - tensor input_479_beta_0_to_fp16 = const()[name = string("input_479_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232525888)))]; - fp16 input_479_epsilon_0_to_fp16 = const()[name = string("input_479_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_479_cast_fp16 = batch_norm(beta = input_479_beta_0_to_fp16, epsilon = input_479_epsilon_0_to_fp16, gamma = input_479_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_177_cast_fp16)[name = string("input_479_cast_fp16")]; - string var_6945_pad_type_0 = const()[name = string("op_6945_pad_type_0"), val = string("valid")]; - tensor var_6945_strides_0 = const()[name = string("op_6945_strides_0"), val = tensor([1, 1])]; - tensor var_6945_pad_0 = const()[name = string("op_6945_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6945_dilations_0 = const()[name = string("op_6945_dilations_0"), val = tensor([1, 1])]; - int32 var_6945_groups_0 = const()[name = string("op_6945_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232528000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234100928))))[name = string("layers_17_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_6945_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6945_dilations_0, groups = var_6945_groups_0, pad = var_6945_pad_0, pad_type = var_6945_pad_type_0, strides = var_6945_strides_0, weight = layers_17_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_479_cast_fp16)[name = string("op_6945_cast_fp16")]; - string var_6951_pad_type_0 = const()[name = string("op_6951_pad_type_0"), val = string("valid")]; - tensor var_6951_strides_0 = const()[name = string("op_6951_strides_0"), val = tensor([1, 1])]; - tensor var_6951_pad_0 = const()[name = string("op_6951_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6951_dilations_0 = const()[name = string("op_6951_dilations_0"), val = tensor([1, 1])]; - int32 var_6951_groups_0 = const()[name = string("op_6951_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234143616))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234105088))))[name = string("layers_17_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6951_cast_fp16 = conv(dilations = var_6951_dilations_0, groups = var_6951_groups_0, pad = var_6951_pad_0, pad_type = var_6951_pad_type_0, strides = var_6951_strides_0, weight = layers_17_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_479_cast_fp16)[name = string("op_6951_cast_fp16")]; - tensor input_481_cast_fp16 = add(x = var_6945_cast_fp16, y = var_6951_cast_fp16)[name = string("input_481_cast_fp16")]; - tensor input_483_cast_fp16 = silu(x = input_481_cast_fp16)[name = string("input_483_cast_fp16")]; - string var_6962_pad_type_0 = const()[name = string("op_6962_pad_type_0"), val = string("valid")]; - tensor var_6962_strides_0 = const()[name = string("op_6962_strides_0"), val = tensor([1, 1])]; - tensor var_6962_pad_0 = const()[name = string("op_6962_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6962_dilations_0 = const()[name = string("op_6962_dilations_0"), val = tensor([1, 1])]; - int32 var_6962_groups_0 = const()[name = string("op_6962_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234667968))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236240896))))[name = string("layers_17_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_6962_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6962_dilations_0, groups = var_6962_groups_0, pad = var_6962_pad_0, pad_type = var_6962_pad_type_0, strides = var_6962_strides_0, weight = layers_17_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_483_cast_fp16)[name = string("op_6962_cast_fp16")]; - string var_6968_pad_type_0 = const()[name = string("op_6968_pad_type_0"), val = string("valid")]; - tensor var_6968_strides_0 = const()[name = string("op_6968_strides_0"), val = tensor([1, 1])]; - tensor var_6968_pad_0 = const()[name = string("op_6968_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_6968_dilations_0 = const()[name = string("op_6968_dilations_0"), val = tensor([1, 1])]; - int32 var_6968_groups_0 = const()[name = string("op_6968_groups_0"), val = int32(1)]; - tensor layers_17_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236287040))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236241984))))[name = string("layers_17_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_6968_cast_fp16 = conv(dilations = var_6968_dilations_0, groups = var_6968_groups_0, pad = var_6968_pad_0, pad_type = var_6968_pad_type_0, strides = var_6968_strides_0, weight = layers_17_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_483_cast_fp16)[name = string("op_6968_cast_fp16")]; - tensor x_109_cast_fp16 = add(x = var_6962_cast_fp16, y = var_6968_cast_fp16)[name = string("x_109_cast_fp16")]; - fp16 var_6970_to_fp16 = const()[name = string("op_6970_to_fp16"), val = fp16(0x1p-1)]; - tensor var_6971_cast_fp16 = mul(x = x_109_cast_fp16, y = var_6970_to_fp16)[name = string("op_6971_cast_fp16")]; - tensor inputs_179_cast_fp16 = add(x = inputs_177_cast_fp16, y = var_6971_cast_fp16)[name = string("inputs_179_cast_fp16")]; - tensor out_179_axes_0 = const()[name = string("out_179_axes_0"), val = tensor([1])]; - fp16 var_6981_to_fp16 = const()[name = string("op_6981_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_179_cast_fp16 = layer_norm(axes = out_179_axes_0, epsilon = var_6981_to_fp16, x = inputs_179_cast_fp16)[name = string("out_179_cast_fp16")]; - tensor inputs_181_gamma_0_to_fp16 = const()[name = string("inputs_181_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236811392)))]; - tensor inputs_181_beta_0_to_fp16 = const()[name = string("inputs_181_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236813504)))]; - fp16 inputs_181_epsilon_0_to_fp16 = const()[name = string("inputs_181_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_181_cast_fp16 = batch_norm(beta = inputs_181_beta_0_to_fp16, epsilon = inputs_181_epsilon_0_to_fp16, gamma = inputs_181_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_179_cast_fp16)[name = string("inputs_181_cast_fp16")]; - int32 var_6995 = const()[name = string("op_6995"), val = int32(3)]; - tensor out_181_axes_0 = const()[name = string("out_181_axes_0"), val = tensor([1])]; - fp16 var_7026_to_fp16 = const()[name = string("op_7026_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_181_cast_fp16 = layer_norm(axes = out_181_axes_0, epsilon = var_7026_to_fp16, x = inputs_181_cast_fp16)[name = string("out_181_cast_fp16")]; - tensor input_485_gamma_0_to_fp16 = const()[name = string("input_485_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236815616)))]; - tensor input_485_beta_0_to_fp16 = const()[name = string("input_485_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236817728)))]; - fp16 input_485_epsilon_0_to_fp16 = const()[name = string("input_485_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_485_cast_fp16 = batch_norm(beta = input_485_beta_0_to_fp16, epsilon = input_485_epsilon_0_to_fp16, gamma = input_485_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_181_cast_fp16)[name = string("input_485_cast_fp16")]; - string var_7046_pad_type_0 = const()[name = string("op_7046_pad_type_0"), val = string("valid")]; - tensor var_7046_strides_0 = const()[name = string("op_7046_strides_0"), val = tensor([1, 1])]; - tensor var_7046_pad_0 = const()[name = string("op_7046_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7046_dilations_0 = const()[name = string("op_7046_dilations_0"), val = tensor([1, 1])]; - int32 var_7046_groups_0 = const()[name = string("op_7046_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236819840))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238392768))))[name = string("layers_18_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_7046_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7046_dilations_0, groups = var_7046_groups_0, pad = var_7046_pad_0, pad_type = var_7046_pad_type_0, strides = var_7046_strides_0, weight = layers_18_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_485_cast_fp16)[name = string("op_7046_cast_fp16")]; - string var_7052_pad_type_0 = const()[name = string("op_7052_pad_type_0"), val = string("valid")]; - tensor var_7052_strides_0 = const()[name = string("op_7052_strides_0"), val = tensor([1, 1])]; - tensor var_7052_pad_0 = const()[name = string("op_7052_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7052_dilations_0 = const()[name = string("op_7052_dilations_0"), val = tensor([1, 1])]; - int32 var_7052_groups_0 = const()[name = string("op_7052_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238437568))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238396928))))[name = string("layers_18_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7052_cast_fp16 = conv(dilations = var_7052_dilations_0, groups = var_7052_groups_0, pad = var_7052_pad_0, pad_type = var_7052_pad_type_0, strides = var_7052_strides_0, weight = layers_18_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_485_cast_fp16)[name = string("op_7052_cast_fp16")]; - tensor input_487_cast_fp16 = add(x = var_7046_cast_fp16, y = var_7052_cast_fp16)[name = string("input_487_cast_fp16")]; - tensor input_489_cast_fp16 = silu(x = input_487_cast_fp16)[name = string("input_489_cast_fp16")]; - string var_7063_pad_type_0 = const()[name = string("op_7063_pad_type_0"), val = string("valid")]; - tensor var_7063_strides_0 = const()[name = string("op_7063_strides_0"), val = tensor([1, 1])]; - tensor var_7063_pad_0 = const()[name = string("op_7063_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7063_dilations_0 = const()[name = string("op_7063_dilations_0"), val = tensor([1, 1])]; - int32 var_7063_groups_0 = const()[name = string("op_7063_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238961920))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240534848))))[name = string("layers_18_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_7063_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7063_dilations_0, groups = var_7063_groups_0, pad = var_7063_pad_0, pad_type = var_7063_pad_type_0, strides = var_7063_strides_0, weight = layers_18_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_489_cast_fp16)[name = string("op_7063_cast_fp16")]; - string var_7069_pad_type_0 = const()[name = string("op_7069_pad_type_0"), val = string("valid")]; - tensor var_7069_strides_0 = const()[name = string("op_7069_strides_0"), val = tensor([1, 1])]; - tensor var_7069_pad_0 = const()[name = string("op_7069_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7069_dilations_0 = const()[name = string("op_7069_dilations_0"), val = tensor([1, 1])]; - int32 var_7069_groups_0 = const()[name = string("op_7069_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240581120))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240535936))))[name = string("layers_18_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7069_cast_fp16 = conv(dilations = var_7069_dilations_0, groups = var_7069_groups_0, pad = var_7069_pad_0, pad_type = var_7069_pad_type_0, strides = var_7069_strides_0, weight = layers_18_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_489_cast_fp16)[name = string("op_7069_cast_fp16")]; - tensor x_111_cast_fp16 = add(x = var_7063_cast_fp16, y = var_7069_cast_fp16)[name = string("x_111_cast_fp16")]; - fp16 var_7071_to_fp16 = const()[name = string("op_7071_to_fp16"), val = fp16(0x1p-1)]; - tensor var_7072_cast_fp16 = mul(x = x_111_cast_fp16, y = var_7071_to_fp16)[name = string("op_7072_cast_fp16")]; - tensor inputs_183_cast_fp16 = add(x = inputs_181_cast_fp16, y = var_7072_cast_fp16)[name = string("inputs_183_cast_fp16")]; - tensor out_183_axes_0 = const()[name = string("out_183_axes_0"), val = tensor([1])]; - fp16 var_7082_to_fp16 = const()[name = string("op_7082_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_183_cast_fp16 = layer_norm(axes = out_183_axes_0, epsilon = var_7082_to_fp16, x = inputs_183_cast_fp16)[name = string("out_183_cast_fp16")]; - tensor obj_75_gamma_0_to_fp16 = const()[name = string("obj_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241105472)))]; - tensor obj_75_beta_0_to_fp16 = const()[name = string("obj_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241107584)))]; - fp16 obj_75_epsilon_0_to_fp16 = const()[name = string("obj_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_75_cast_fp16 = batch_norm(beta = obj_75_beta_0_to_fp16, epsilon = obj_75_epsilon_0_to_fp16, gamma = obj_75_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_183_cast_fp16)[name = string("obj_75_cast_fp16")]; - string var_7107_pad_type_0 = const()[name = string("op_7107_pad_type_0"), val = string("valid")]; - tensor var_7107_strides_0 = const()[name = string("op_7107_strides_0"), val = tensor([1, 1])]; - tensor var_7107_pad_0 = const()[name = string("op_7107_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7107_dilations_0 = const()[name = string("op_7107_dilations_0"), val = tensor([1, 1])]; - int32 var_7107_groups_0 = const()[name = string("op_7107_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241109696))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241502976))))[name = string("layers_18_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7107_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7107_dilations_0, groups = var_7107_groups_0, pad = var_7107_pad_0, pad_type = var_7107_pad_type_0, strides = var_7107_strides_0, weight = layers_18_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_75_cast_fp16)[name = string("op_7107_cast_fp16")]; - string var_7113_pad_type_0 = const()[name = string("op_7113_pad_type_0"), val = string("valid")]; - tensor var_7113_strides_0 = const()[name = string("op_7113_strides_0"), val = tensor([1, 1])]; - tensor var_7113_pad_0 = const()[name = string("op_7113_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7113_dilations_0 = const()[name = string("op_7113_dilations_0"), val = tensor([1, 1])]; - int32 var_7113_groups_0 = const()[name = string("op_7113_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241515456))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241504064))))[name = string("layers_18_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7113_cast_fp16 = conv(dilations = var_7113_dilations_0, groups = var_7113_groups_0, pad = var_7113_pad_0, pad_type = var_7113_pad_type_0, strides = var_7113_strides_0, weight = layers_18_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_75_cast_fp16)[name = string("op_7113_cast_fp16")]; - tensor query_73_cast_fp16 = add(x = var_7107_cast_fp16, y = var_7113_cast_fp16)[name = string("query_73_cast_fp16")]; - string var_7122_pad_type_0 = const()[name = string("op_7122_pad_type_0"), val = string("valid")]; - tensor var_7122_strides_0 = const()[name = string("op_7122_strides_0"), val = tensor([1, 1])]; - tensor var_7122_pad_0 = const()[name = string("op_7122_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7122_dilations_0 = const()[name = string("op_7122_dilations_0"), val = tensor([1, 1])]; - int32 var_7122_groups_0 = const()[name = string("op_7122_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241646592))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242039872))))[name = string("layers_18_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7122_cast_fp16 = conv(dilations = var_7122_dilations_0, groups = var_7122_groups_0, pad = var_7122_pad_0, pad_type = var_7122_pad_type_0, strides = var_7122_strides_0, weight = layers_18_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_75_cast_fp16)[name = string("op_7122_cast_fp16")]; - string var_7128_pad_type_0 = const()[name = string("op_7128_pad_type_0"), val = string("valid")]; - tensor var_7128_strides_0 = const()[name = string("op_7128_strides_0"), val = tensor([1, 1])]; - tensor var_7128_pad_0 = const()[name = string("op_7128_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7128_dilations_0 = const()[name = string("op_7128_dilations_0"), val = tensor([1, 1])]; - int32 var_7128_groups_0 = const()[name = string("op_7128_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242056256))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242040960))))[name = string("layers_18_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7128_cast_fp16 = conv(dilations = var_7128_dilations_0, groups = var_7128_groups_0, pad = var_7128_pad_0, pad_type = var_7128_pad_type_0, strides = var_7128_strides_0, weight = layers_18_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_75_cast_fp16)[name = string("op_7128_cast_fp16")]; - tensor key_37_cast_fp16 = add(x = var_7122_cast_fp16, y = var_7128_cast_fp16)[name = string("key_37_cast_fp16")]; - string var_7138_pad_type_0 = const()[name = string("op_7138_pad_type_0"), val = string("valid")]; - tensor var_7138_strides_0 = const()[name = string("op_7138_strides_0"), val = tensor([1, 1])]; - tensor var_7138_pad_0 = const()[name = string("op_7138_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7138_dilations_0 = const()[name = string("op_7138_dilations_0"), val = tensor([1, 1])]; - int32 var_7138_groups_0 = const()[name = string("op_7138_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242187392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242580672))))[name = string("layers_18_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7138_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7138_dilations_0, groups = var_7138_groups_0, pad = var_7138_pad_0, pad_type = var_7138_pad_type_0, strides = var_7138_strides_0, weight = layers_18_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_75_cast_fp16)[name = string("op_7138_cast_fp16")]; - string var_7144_pad_type_0 = const()[name = string("op_7144_pad_type_0"), val = string("valid")]; - tensor var_7144_strides_0 = const()[name = string("op_7144_strides_0"), val = tensor([1, 1])]; - tensor var_7144_pad_0 = const()[name = string("op_7144_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7144_dilations_0 = const()[name = string("op_7144_dilations_0"), val = tensor([1, 1])]; - int32 var_7144_groups_0 = const()[name = string("op_7144_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242589824))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242581760))))[name = string("layers_18_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7144_cast_fp16 = conv(dilations = var_7144_dilations_0, groups = var_7144_groups_0, pad = var_7144_pad_0, pad_type = var_7144_pad_type_0, strides = var_7144_strides_0, weight = layers_18_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_75_cast_fp16)[name = string("op_7144_cast_fp16")]; - tensor value_37_cast_fp16 = add(x = var_7138_cast_fp16, y = var_7144_cast_fp16)[name = string("value_37_cast_fp16")]; - tensor var_7147_to_fp16 = const()[name = string("op_7147_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242720960)))]; - tensor query_75_cast_fp16 = add(x = query_73_cast_fp16, y = var_7147_to_fp16)[name = string("query_75_cast_fp16")]; - tensor var_7150_to_fp16 = const()[name = string("op_7150_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242723072)))]; - tensor q_with_bias_v_37_cast_fp16 = add(x = query_73_cast_fp16, y = var_7150_to_fp16)[name = string("q_with_bias_v_37_cast_fp16")]; - string var_7160_pad_type_0 = const()[name = string("op_7160_pad_type_0"), val = string("valid")]; - tensor var_7160_strides_0 = const()[name = string("op_7160_strides_0"), val = tensor([1, 1])]; - tensor var_7160_pad_0 = const()[name = string("op_7160_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7160_dilations_0 = const()[name = string("op_7160_dilations_0"), val = tensor([1, 1])]; - int32 var_7160_groups_0 = const()[name = string("op_7160_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242725184))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243118464))))[name = string("layers_18_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_7160_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7160_dilations_0, groups = var_7160_groups_0, pad = var_7160_pad_0, pad_type = var_7160_pad_type_0, strides = var_7160_strides_0, weight = layers_18_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_7160_cast_fp16")]; - string var_7166_pad_type_0 = const()[name = string("op_7166_pad_type_0"), val = string("valid")]; - tensor var_7166_strides_0 = const()[name = string("op_7166_strides_0"), val = tensor([1, 1])]; - tensor var_7166_pad_0 = const()[name = string("op_7166_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7166_dilations_0 = const()[name = string("op_7166_dilations_0"), val = tensor([1, 1])]; - int32 var_7166_groups_0 = const()[name = string("op_7166_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243148928))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243119552))))[name = string("layers_18_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7166_cast_fp16 = conv(dilations = var_7166_dilations_0, groups = var_7166_groups_0, pad = var_7166_pad_0, pad_type = var_7166_pad_type_0, strides = var_7166_strides_0, weight = layers_18_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_7166_cast_fp16")]; - tensor p_37_cast_fp16 = add(x = var_7160_cast_fp16, y = var_7166_cast_fp16)[name = string("p_37_cast_fp16")]; - tensor var_7170 = const()[name = string("op_7170"), val = tensor([1, 8, 128, 188])]; - tensor var_7171_cast_fp16 = reshape(shape = var_7170, x = q_with_bias_v_37_cast_fp16)[name = string("op_7171_cast_fp16")]; - tensor var_7172 = const()[name = string("op_7172"), val = tensor([1, 8, 128, -1])]; - tensor var_7173_cast_fp16 = reshape(shape = var_7172, x = p_37_cast_fp16)[name = string("op_7173_cast_fp16")]; - bool matrix_bd_145_transpose_x_0 = const()[name = string("matrix_bd_145_transpose_x_0"), val = bool(true)]; - bool matrix_bd_145_transpose_y_0 = const()[name = string("matrix_bd_145_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_145_cast_fp16 = matmul(transpose_x = matrix_bd_145_transpose_x_0, transpose_y = matrix_bd_145_transpose_y_0, x = var_7171_cast_fp16, y = var_7173_cast_fp16)[name = string("matrix_bd_145_cast_fp16")]; - tensor matrix_bd_147_pad_0 = const()[name = string("matrix_bd_147_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_147_mode_0 = const()[name = string("matrix_bd_147_mode_0"), val = string("constant")]; - fp16 const_208_to_fp16 = const()[name = string("const_208_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_147_cast_fp16 = pad(constant_val = const_208_to_fp16, mode = matrix_bd_147_mode_0, pad = matrix_bd_147_pad_0, x = matrix_bd_145_cast_fp16)[name = string("matrix_bd_147_cast_fp16")]; - tensor var_7182 = const()[name = string("op_7182"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_149_cast_fp16 = reshape(shape = var_7182, x = matrix_bd_147_cast_fp16)[name = string("matrix_bd_149_cast_fp16")]; - tensor var_7186_begin_0 = const()[name = string("op_7186_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_7186_end_0 = const()[name = string("op_7186_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_7186_end_mask_0 = const()[name = string("op_7186_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_7186_cast_fp16 = slice_by_index(begin = var_7186_begin_0, end = var_7186_end_0, end_mask = var_7186_end_mask_0, x = matrix_bd_149_cast_fp16)[name = string("op_7186_cast_fp16")]; - tensor var_7187 = const()[name = string("op_7187"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_151_cast_fp16 = reshape(shape = var_7187, x = var_7186_cast_fp16)[name = string("matrix_bd_151_cast_fp16")]; - tensor var_7192_begin_0 = const()[name = string("op_7192_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7192_end_0 = const()[name = string("op_7192_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_7192_end_mask_0 = const()[name = string("op_7192_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_7192_cast_fp16 = slice_by_index(begin = var_7192_begin_0, end = var_7192_end_0, end_mask = var_7192_end_mask_0, x = matrix_bd_151_cast_fp16)[name = string("op_7192_cast_fp16")]; - fp16 var_7193_to_fp16 = const()[name = string("op_7193_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_37_cast_fp16 = mul(x = var_7192_cast_fp16, y = var_7193_to_fp16)[name = string("qk_mask_37_cast_fp16")]; - tensor var_7197 = const()[name = string("op_7197"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_37_cast_fp16 = reshape(shape = var_7197, x = query_75_cast_fp16)[name = string("mh_q_37_cast_fp16")]; - fp16 var_7199_to_fp16 = const()[name = string("op_7199_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_7200_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_7199_to_fp16)[name = string("op_7200_cast_fp16")]; - tensor var_7203 = const()[name = string("op_7203"), val = tensor([1, 8, 128, 188])]; - tensor var_7204_cast_fp16 = reshape(shape = var_7203, x = key_37_cast_fp16)[name = string("op_7204_cast_fp16")]; - bool mh_w_73_transpose_x_0 = const()[name = string("mh_w_73_transpose_x_0"), val = bool(true)]; - bool mh_w_73_transpose_y_0 = const()[name = string("mh_w_73_transpose_y_0"), val = bool(false)]; - tensor mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_7200_cast_fp16, y = var_7204_cast_fp16)[name = string("mh_w_73_cast_fp16")]; - tensor mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = qk_mask_37_cast_fp16)[name = string("mh_w_75_cast_fp16")]; - tensor var_7208_cast_fp16 = softmax(axis = var_6995, x = mh_w_75_cast_fp16)[name = string("op_7208_cast_fp16")]; - tensor var_7209 = const()[name = string("op_7209"), val = tensor([1, 8, 128, 188])]; - tensor var_7210_cast_fp16 = reshape(shape = var_7209, x = value_37_cast_fp16)[name = string("op_7210_cast_fp16")]; - bool attn_37_transpose_x_0 = const()[name = string("attn_37_transpose_x_0"), val = bool(false)]; - bool attn_37_transpose_y_0 = const()[name = string("attn_37_transpose_y_0"), val = bool(true)]; - tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_7210_cast_fp16, y = var_7208_cast_fp16)[name = string("attn_37_cast_fp16")]; - tensor var_7213 = const()[name = string("op_7213"), val = tensor([1, 1024, 1, 188])]; - tensor input_491_cast_fp16 = reshape(shape = var_7213, x = attn_37_cast_fp16)[name = string("input_491_cast_fp16")]; - string var_7223_pad_type_0 = const()[name = string("op_7223_pad_type_0"), val = string("valid")]; - tensor var_7223_strides_0 = const()[name = string("op_7223_strides_0"), val = tensor([1, 1])]; - tensor var_7223_pad_0 = const()[name = string("op_7223_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7223_dilations_0 = const()[name = string("op_7223_dilations_0"), val = tensor([1, 1])]; - int32 var_7223_groups_0 = const()[name = string("op_7223_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243280064))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243673344))))[name = string("layers_18_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7223_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7223_dilations_0, groups = var_7223_groups_0, pad = var_7223_pad_0, pad_type = var_7223_pad_type_0, strides = var_7223_strides_0, weight = layers_18_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_491_cast_fp16)[name = string("op_7223_cast_fp16")]; - string var_7229_pad_type_0 = const()[name = string("op_7229_pad_type_0"), val = string("valid")]; - tensor var_7229_strides_0 = const()[name = string("op_7229_strides_0"), val = tensor([1, 1])]; - tensor var_7229_pad_0 = const()[name = string("op_7229_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7229_dilations_0 = const()[name = string("op_7229_dilations_0"), val = tensor([1, 1])]; - int32 var_7229_groups_0 = const()[name = string("op_7229_groups_0"), val = int32(1)]; - tensor layers_18_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243683328))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243674432))))[name = string("layers_18_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7229_cast_fp16 = conv(dilations = var_7229_dilations_0, groups = var_7229_groups_0, pad = var_7229_pad_0, pad_type = var_7229_pad_type_0, strides = var_7229_strides_0, weight = layers_18_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_491_cast_fp16)[name = string("op_7229_cast_fp16")]; - tensor obj_77_cast_fp16 = add(x = var_7223_cast_fp16, y = var_7229_cast_fp16)[name = string("obj_77_cast_fp16")]; - tensor inputs_185_cast_fp16 = add(x = inputs_183_cast_fp16, y = obj_77_cast_fp16)[name = string("inputs_185_cast_fp16")]; - tensor out_185_axes_0 = const()[name = string("out_185_axes_0"), val = tensor([1])]; - fp16 var_7240_to_fp16 = const()[name = string("op_7240_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_185_cast_fp16 = layer_norm(axes = out_185_axes_0, epsilon = var_7240_to_fp16, x = inputs_185_cast_fp16)[name = string("out_185_cast_fp16")]; - tensor input_493_gamma_0_to_fp16 = const()[name = string("input_493_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243814464)))]; - tensor input_493_beta_0_to_fp16 = const()[name = string("input_493_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243816576)))]; - fp16 input_493_epsilon_0_to_fp16 = const()[name = string("input_493_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_493_cast_fp16 = batch_norm(beta = input_493_beta_0_to_fp16, epsilon = input_493_epsilon_0_to_fp16, gamma = input_493_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_185_cast_fp16)[name = string("input_493_cast_fp16")]; - string var_7261_pad_type_0 = const()[name = string("op_7261_pad_type_0"), val = string("valid")]; - tensor var_7261_strides_0 = const()[name = string("op_7261_strides_0"), val = tensor([1, 1])]; - tensor var_7261_pad_0 = const()[name = string("op_7261_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7261_dilations_0 = const()[name = string("op_7261_dilations_0"), val = tensor([1, 1])]; - int32 var_7261_groups_0 = const()[name = string("op_7261_groups_0"), val = int32(1)]; - tensor layers_18_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243818688))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244605184))))[name = string("layers_18_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_7261_cast_fp16 = conv(dilations = var_7261_dilations_0, groups = var_7261_groups_0, pad = var_7261_pad_0, pad_type = var_7261_pad_type_0, strides = var_7261_strides_0, weight = layers_18_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_493_cast_fp16)[name = string("op_7261_cast_fp16")]; - string var_7267_pad_type_0 = const()[name = string("op_7267_pad_type_0"), val = string("valid")]; - tensor var_7267_strides_0 = const()[name = string("op_7267_strides_0"), val = tensor([1, 1])]; - tensor var_7267_pad_0 = const()[name = string("op_7267_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7267_dilations_0 = const()[name = string("op_7267_dilations_0"), val = tensor([1, 1])]; - int32 var_7267_groups_0 = const()[name = string("op_7267_groups_0"), val = int32(1)]; - tensor layers_18_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244629504))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244607296))))[name = string("layers_18_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7267_cast_fp16 = conv(dilations = var_7267_dilations_0, groups = var_7267_groups_0, pad = var_7267_pad_0, pad_type = var_7267_pad_type_0, strides = var_7267_strides_0, weight = layers_18_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_493_cast_fp16)[name = string("op_7267_cast_fp16")]; - tensor input_495_cast_fp16 = add(x = var_7261_cast_fp16, y = var_7267_cast_fp16)[name = string("input_495_cast_fp16")]; - int32 input_497_split_num_splits_0 = const()[name = string("input_497_split_num_splits_0"), val = int32(2)]; - int32 input_497_split_axis_0 = const()[name = string("input_497_split_axis_0"), val = int32(1)]; - tensor input_497_split_cast_fp16_0, tensor input_497_split_cast_fp16_1 = split(axis = input_497_split_axis_0, num_splits = input_497_split_num_splits_0, x = input_495_cast_fp16)[name = string("input_497_split_cast_fp16")]; - tensor input_497_split_1_sigmoid_cast_fp16 = sigmoid(x = input_497_split_cast_fp16_1)[name = string("input_497_split_1_sigmoid_cast_fp16")]; - tensor input_497_cast_fp16 = mul(x = input_497_split_cast_fp16_0, y = input_497_split_1_sigmoid_cast_fp16)[name = string("input_497_cast_fp16")]; - string input_499_pad_type_0 = const()[name = string("input_499_pad_type_0"), val = string("custom")]; - tensor input_499_pad_0 = const()[name = string("input_499_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_499_groups_0 = const()[name = string("input_499_groups_0"), val = int32(1024)]; - tensor input_499_strides_0 = const()[name = string("input_499_strides_0"), val = tensor([1, 1])]; - tensor input_499_dilations_0 = const()[name = string("input_499_dilations_0"), val = tensor([1, 1])]; - tensor const_304_to_fp16 = const()[name = string("const_304_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244891712)))]; - tensor const_305_to_fp16 = const()[name = string("const_305_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244910208)))]; - tensor input_501_cast_fp16 = conv(bias = const_305_to_fp16, dilations = input_499_dilations_0, groups = input_499_groups_0, pad = input_499_pad_0, pad_type = input_499_pad_type_0, strides = input_499_strides_0, weight = const_304_to_fp16, x = input_497_cast_fp16)[name = string("input_501_cast_fp16")]; - tensor input_503_cast_fp16 = silu(x = input_501_cast_fp16)[name = string("input_503_cast_fp16")]; - string var_7289_pad_type_0 = const()[name = string("op_7289_pad_type_0"), val = string("valid")]; - tensor var_7289_strides_0 = const()[name = string("op_7289_strides_0"), val = tensor([1, 1])]; - tensor var_7289_pad_0 = const()[name = string("op_7289_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7289_dilations_0 = const()[name = string("op_7289_dilations_0"), val = tensor([1, 1])]; - int32 var_7289_groups_0 = const()[name = string("op_7289_groups_0"), val = int32(1)]; - tensor layers_18_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244912320))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245305600))))[name = string("layers_18_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_7289_cast_fp16 = conv(dilations = var_7289_dilations_0, groups = var_7289_groups_0, pad = var_7289_pad_0, pad_type = var_7289_pad_type_0, strides = var_7289_strides_0, weight = layers_18_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_503_cast_fp16)[name = string("op_7289_cast_fp16")]; - string var_7295_pad_type_0 = const()[name = string("op_7295_pad_type_0"), val = string("valid")]; - tensor var_7295_strides_0 = const()[name = string("op_7295_strides_0"), val = tensor([1, 1])]; - tensor var_7295_pad_0 = const()[name = string("op_7295_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7295_dilations_0 = const()[name = string("op_7295_dilations_0"), val = tensor([1, 1])]; - int32 var_7295_groups_0 = const()[name = string("op_7295_groups_0"), val = int32(1)]; - tensor layers_18_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245315392))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245306688))))[name = string("layers_18_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7295_cast_fp16 = conv(dilations = var_7295_dilations_0, groups = var_7295_groups_0, pad = var_7295_pad_0, pad_type = var_7295_pad_type_0, strides = var_7295_strides_0, weight = layers_18_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_503_cast_fp16)[name = string("op_7295_cast_fp16")]; - tensor x_113_cast_fp16 = add(x = var_7289_cast_fp16, y = var_7295_cast_fp16)[name = string("x_113_cast_fp16")]; - tensor inputs_187_cast_fp16 = add(x = inputs_185_cast_fp16, y = x_113_cast_fp16)[name = string("inputs_187_cast_fp16")]; - tensor out_187_axes_0 = const()[name = string("out_187_axes_0"), val = tensor([1])]; - fp16 var_7306_to_fp16 = const()[name = string("op_7306_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_187_cast_fp16 = layer_norm(axes = out_187_axes_0, epsilon = var_7306_to_fp16, x = inputs_187_cast_fp16)[name = string("out_187_cast_fp16")]; - tensor input_505_gamma_0_to_fp16 = const()[name = string("input_505_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245446528)))]; - tensor input_505_beta_0_to_fp16 = const()[name = string("input_505_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245448640)))]; - fp16 input_505_epsilon_0_to_fp16 = const()[name = string("input_505_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_505_cast_fp16 = batch_norm(beta = input_505_beta_0_to_fp16, epsilon = input_505_epsilon_0_to_fp16, gamma = input_505_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_187_cast_fp16)[name = string("input_505_cast_fp16")]; - string var_7326_pad_type_0 = const()[name = string("op_7326_pad_type_0"), val = string("valid")]; - tensor var_7326_strides_0 = const()[name = string("op_7326_strides_0"), val = tensor([1, 1])]; - tensor var_7326_pad_0 = const()[name = string("op_7326_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7326_dilations_0 = const()[name = string("op_7326_dilations_0"), val = tensor([1, 1])]; - int32 var_7326_groups_0 = const()[name = string("op_7326_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245450752))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247023680))))[name = string("layers_18_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_7326_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7326_dilations_0, groups = var_7326_groups_0, pad = var_7326_pad_0, pad_type = var_7326_pad_type_0, strides = var_7326_strides_0, weight = layers_18_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_505_cast_fp16)[name = string("op_7326_cast_fp16")]; - string var_7332_pad_type_0 = const()[name = string("op_7332_pad_type_0"), val = string("valid")]; - tensor var_7332_strides_0 = const()[name = string("op_7332_strides_0"), val = tensor([1, 1])]; - tensor var_7332_pad_0 = const()[name = string("op_7332_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7332_dilations_0 = const()[name = string("op_7332_dilations_0"), val = tensor([1, 1])]; - int32 var_7332_groups_0 = const()[name = string("op_7332_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247064128))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247027840))))[name = string("layers_18_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7332_cast_fp16 = conv(dilations = var_7332_dilations_0, groups = var_7332_groups_0, pad = var_7332_pad_0, pad_type = var_7332_pad_type_0, strides = var_7332_strides_0, weight = layers_18_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_505_cast_fp16)[name = string("op_7332_cast_fp16")]; - tensor input_507_cast_fp16 = add(x = var_7326_cast_fp16, y = var_7332_cast_fp16)[name = string("input_507_cast_fp16")]; - tensor input_509_cast_fp16 = silu(x = input_507_cast_fp16)[name = string("input_509_cast_fp16")]; - string var_7343_pad_type_0 = const()[name = string("op_7343_pad_type_0"), val = string("valid")]; - tensor var_7343_strides_0 = const()[name = string("op_7343_strides_0"), val = tensor([1, 1])]; - tensor var_7343_pad_0 = const()[name = string("op_7343_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7343_dilations_0 = const()[name = string("op_7343_dilations_0"), val = tensor([1, 1])]; - int32 var_7343_groups_0 = const()[name = string("op_7343_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247588480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249161408))))[name = string("layers_18_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_7343_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7343_dilations_0, groups = var_7343_groups_0, pad = var_7343_pad_0, pad_type = var_7343_pad_type_0, strides = var_7343_strides_0, weight = layers_18_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_509_cast_fp16)[name = string("op_7343_cast_fp16")]; - string var_7349_pad_type_0 = const()[name = string("op_7349_pad_type_0"), val = string("valid")]; - tensor var_7349_strides_0 = const()[name = string("op_7349_strides_0"), val = tensor([1, 1])]; - tensor var_7349_pad_0 = const()[name = string("op_7349_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7349_dilations_0 = const()[name = string("op_7349_dilations_0"), val = tensor([1, 1])]; - int32 var_7349_groups_0 = const()[name = string("op_7349_groups_0"), val = int32(1)]; - tensor layers_18_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249205248))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249162496))))[name = string("layers_18_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7349_cast_fp16 = conv(dilations = var_7349_dilations_0, groups = var_7349_groups_0, pad = var_7349_pad_0, pad_type = var_7349_pad_type_0, strides = var_7349_strides_0, weight = layers_18_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_509_cast_fp16)[name = string("op_7349_cast_fp16")]; - tensor x_115_cast_fp16 = add(x = var_7343_cast_fp16, y = var_7349_cast_fp16)[name = string("x_115_cast_fp16")]; - fp16 var_7351_to_fp16 = const()[name = string("op_7351_to_fp16"), val = fp16(0x1p-1)]; - tensor var_7352_cast_fp16 = mul(x = x_115_cast_fp16, y = var_7351_to_fp16)[name = string("op_7352_cast_fp16")]; - tensor inputs_189_cast_fp16 = add(x = inputs_187_cast_fp16, y = var_7352_cast_fp16)[name = string("inputs_189_cast_fp16")]; - tensor out_189_axes_0 = const()[name = string("out_189_axes_0"), val = tensor([1])]; - fp16 var_7362_to_fp16 = const()[name = string("op_7362_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_189_cast_fp16 = layer_norm(axes = out_189_axes_0, epsilon = var_7362_to_fp16, x = inputs_189_cast_fp16)[name = string("out_189_cast_fp16")]; - tensor inputs_191_gamma_0_to_fp16 = const()[name = string("inputs_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249729600)))]; - tensor inputs_191_beta_0_to_fp16 = const()[name = string("inputs_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249731712)))]; - fp16 inputs_191_epsilon_0_to_fp16 = const()[name = string("inputs_191_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_191_cast_fp16 = batch_norm(beta = inputs_191_beta_0_to_fp16, epsilon = inputs_191_epsilon_0_to_fp16, gamma = inputs_191_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_189_cast_fp16)[name = string("inputs_191_cast_fp16")]; - int32 var_7376 = const()[name = string("op_7376"), val = int32(3)]; - tensor out_191_axes_0 = const()[name = string("out_191_axes_0"), val = tensor([1])]; - fp16 var_7407_to_fp16 = const()[name = string("op_7407_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_191_cast_fp16 = layer_norm(axes = out_191_axes_0, epsilon = var_7407_to_fp16, x = inputs_191_cast_fp16)[name = string("out_191_cast_fp16")]; - tensor input_511_gamma_0_to_fp16 = const()[name = string("input_511_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249733824)))]; - tensor input_511_beta_0_to_fp16 = const()[name = string("input_511_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249735936)))]; - fp16 input_511_epsilon_0_to_fp16 = const()[name = string("input_511_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_511_cast_fp16 = batch_norm(beta = input_511_beta_0_to_fp16, epsilon = input_511_epsilon_0_to_fp16, gamma = input_511_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_191_cast_fp16)[name = string("input_511_cast_fp16")]; - string var_7427_pad_type_0 = const()[name = string("op_7427_pad_type_0"), val = string("valid")]; - tensor var_7427_strides_0 = const()[name = string("op_7427_strides_0"), val = tensor([1, 1])]; - tensor var_7427_pad_0 = const()[name = string("op_7427_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7427_dilations_0 = const()[name = string("op_7427_dilations_0"), val = tensor([1, 1])]; - int32 var_7427_groups_0 = const()[name = string("op_7427_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249738048))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251310976))))[name = string("layers_19_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_7427_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7427_dilations_0, groups = var_7427_groups_0, pad = var_7427_pad_0, pad_type = var_7427_pad_type_0, strides = var_7427_strides_0, weight = layers_19_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_511_cast_fp16)[name = string("op_7427_cast_fp16")]; - string var_7433_pad_type_0 = const()[name = string("op_7433_pad_type_0"), val = string("valid")]; - tensor var_7433_strides_0 = const()[name = string("op_7433_strides_0"), val = tensor([1, 1])]; - tensor var_7433_pad_0 = const()[name = string("op_7433_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7433_dilations_0 = const()[name = string("op_7433_dilations_0"), val = tensor([1, 1])]; - int32 var_7433_groups_0 = const()[name = string("op_7433_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251352448))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251315136))))[name = string("layers_19_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7433_cast_fp16 = conv(dilations = var_7433_dilations_0, groups = var_7433_groups_0, pad = var_7433_pad_0, pad_type = var_7433_pad_type_0, strides = var_7433_strides_0, weight = layers_19_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_511_cast_fp16)[name = string("op_7433_cast_fp16")]; - tensor input_513_cast_fp16 = add(x = var_7427_cast_fp16, y = var_7433_cast_fp16)[name = string("input_513_cast_fp16")]; - tensor input_515_cast_fp16 = silu(x = input_513_cast_fp16)[name = string("input_515_cast_fp16")]; - string var_7444_pad_type_0 = const()[name = string("op_7444_pad_type_0"), val = string("valid")]; - tensor var_7444_strides_0 = const()[name = string("op_7444_strides_0"), val = tensor([1, 1])]; - tensor var_7444_pad_0 = const()[name = string("op_7444_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7444_dilations_0 = const()[name = string("op_7444_dilations_0"), val = tensor([1, 1])]; - int32 var_7444_groups_0 = const()[name = string("op_7444_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251876800))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253449728))))[name = string("layers_19_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_7444_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7444_dilations_0, groups = var_7444_groups_0, pad = var_7444_pad_0, pad_type = var_7444_pad_type_0, strides = var_7444_strides_0, weight = layers_19_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_515_cast_fp16)[name = string("op_7444_cast_fp16")]; - string var_7450_pad_type_0 = const()[name = string("op_7450_pad_type_0"), val = string("valid")]; - tensor var_7450_strides_0 = const()[name = string("op_7450_strides_0"), val = tensor([1, 1])]; - tensor var_7450_pad_0 = const()[name = string("op_7450_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7450_dilations_0 = const()[name = string("op_7450_dilations_0"), val = tensor([1, 1])]; - int32 var_7450_groups_0 = const()[name = string("op_7450_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253495168))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253450816))))[name = string("layers_19_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7450_cast_fp16 = conv(dilations = var_7450_dilations_0, groups = var_7450_groups_0, pad = var_7450_pad_0, pad_type = var_7450_pad_type_0, strides = var_7450_strides_0, weight = layers_19_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_515_cast_fp16)[name = string("op_7450_cast_fp16")]; - tensor x_117_cast_fp16 = add(x = var_7444_cast_fp16, y = var_7450_cast_fp16)[name = string("x_117_cast_fp16")]; - fp16 var_7452_to_fp16 = const()[name = string("op_7452_to_fp16"), val = fp16(0x1p-1)]; - tensor var_7453_cast_fp16 = mul(x = x_117_cast_fp16, y = var_7452_to_fp16)[name = string("op_7453_cast_fp16")]; - tensor inputs_193_cast_fp16 = add(x = inputs_191_cast_fp16, y = var_7453_cast_fp16)[name = string("inputs_193_cast_fp16")]; - tensor out_193_axes_0 = const()[name = string("out_193_axes_0"), val = tensor([1])]; - fp16 var_7463_to_fp16 = const()[name = string("op_7463_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_193_cast_fp16 = layer_norm(axes = out_193_axes_0, epsilon = var_7463_to_fp16, x = inputs_193_cast_fp16)[name = string("out_193_cast_fp16")]; - tensor obj_79_gamma_0_to_fp16 = const()[name = string("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254019520)))]; - tensor obj_79_beta_0_to_fp16 = const()[name = string("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254021632)))]; - fp16 obj_79_epsilon_0_to_fp16 = const()[name = string("obj_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_193_cast_fp16)[name = string("obj_79_cast_fp16")]; - string var_7488_pad_type_0 = const()[name = string("op_7488_pad_type_0"), val = string("valid")]; - tensor var_7488_strides_0 = const()[name = string("op_7488_strides_0"), val = tensor([1, 1])]; - tensor var_7488_pad_0 = const()[name = string("op_7488_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7488_dilations_0 = const()[name = string("op_7488_dilations_0"), val = tensor([1, 1])]; - int32 var_7488_groups_0 = const()[name = string("op_7488_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254023744))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254417024))))[name = string("layers_19_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7488_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7488_dilations_0, groups = var_7488_groups_0, pad = var_7488_pad_0, pad_type = var_7488_pad_type_0, strides = var_7488_strides_0, weight = layers_19_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = string("op_7488_cast_fp16")]; - string var_7494_pad_type_0 = const()[name = string("op_7494_pad_type_0"), val = string("valid")]; - tensor var_7494_strides_0 = const()[name = string("op_7494_strides_0"), val = tensor([1, 1])]; - tensor var_7494_pad_0 = const()[name = string("op_7494_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7494_dilations_0 = const()[name = string("op_7494_dilations_0"), val = tensor([1, 1])]; - int32 var_7494_groups_0 = const()[name = string("op_7494_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254427072))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254418112))))[name = string("layers_19_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7494_cast_fp16 = conv(dilations = var_7494_dilations_0, groups = var_7494_groups_0, pad = var_7494_pad_0, pad_type = var_7494_pad_type_0, strides = var_7494_strides_0, weight = layers_19_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_79_cast_fp16)[name = string("op_7494_cast_fp16")]; - tensor query_77_cast_fp16 = add(x = var_7488_cast_fp16, y = var_7494_cast_fp16)[name = string("query_77_cast_fp16")]; - string var_7503_pad_type_0 = const()[name = string("op_7503_pad_type_0"), val = string("valid")]; - tensor var_7503_strides_0 = const()[name = string("op_7503_strides_0"), val = tensor([1, 1])]; - tensor var_7503_pad_0 = const()[name = string("op_7503_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7503_dilations_0 = const()[name = string("op_7503_dilations_0"), val = tensor([1, 1])]; - int32 var_7503_groups_0 = const()[name = string("op_7503_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254558208))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254951488))))[name = string("layers_19_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7503_cast_fp16 = conv(dilations = var_7503_dilations_0, groups = var_7503_groups_0, pad = var_7503_pad_0, pad_type = var_7503_pad_type_0, strides = var_7503_strides_0, weight = layers_19_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = string("op_7503_cast_fp16")]; - string var_7509_pad_type_0 = const()[name = string("op_7509_pad_type_0"), val = string("valid")]; - tensor var_7509_strides_0 = const()[name = string("op_7509_strides_0"), val = tensor([1, 1])]; - tensor var_7509_pad_0 = const()[name = string("op_7509_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7509_dilations_0 = const()[name = string("op_7509_dilations_0"), val = tensor([1, 1])]; - int32 var_7509_groups_0 = const()[name = string("op_7509_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254963648))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254952576))))[name = string("layers_19_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7509_cast_fp16 = conv(dilations = var_7509_dilations_0, groups = var_7509_groups_0, pad = var_7509_pad_0, pad_type = var_7509_pad_type_0, strides = var_7509_strides_0, weight = layers_19_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_79_cast_fp16)[name = string("op_7509_cast_fp16")]; - tensor key_39_cast_fp16 = add(x = var_7503_cast_fp16, y = var_7509_cast_fp16)[name = string("key_39_cast_fp16")]; - string var_7519_pad_type_0 = const()[name = string("op_7519_pad_type_0"), val = string("valid")]; - tensor var_7519_strides_0 = const()[name = string("op_7519_strides_0"), val = tensor([1, 1])]; - tensor var_7519_pad_0 = const()[name = string("op_7519_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7519_dilations_0 = const()[name = string("op_7519_dilations_0"), val = tensor([1, 1])]; - int32 var_7519_groups_0 = const()[name = string("op_7519_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255094784))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255488064))))[name = string("layers_19_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7519_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7519_dilations_0, groups = var_7519_groups_0, pad = var_7519_pad_0, pad_type = var_7519_pad_type_0, strides = var_7519_strides_0, weight = layers_19_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = string("op_7519_cast_fp16")]; - string var_7525_pad_type_0 = const()[name = string("op_7525_pad_type_0"), val = string("valid")]; - tensor var_7525_strides_0 = const()[name = string("op_7525_strides_0"), val = tensor([1, 1])]; - tensor var_7525_pad_0 = const()[name = string("op_7525_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7525_dilations_0 = const()[name = string("op_7525_dilations_0"), val = tensor([1, 1])]; - int32 var_7525_groups_0 = const()[name = string("op_7525_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255496384))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255489152))))[name = string("layers_19_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7525_cast_fp16 = conv(dilations = var_7525_dilations_0, groups = var_7525_groups_0, pad = var_7525_pad_0, pad_type = var_7525_pad_type_0, strides = var_7525_strides_0, weight = layers_19_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_79_cast_fp16)[name = string("op_7525_cast_fp16")]; - tensor value_39_cast_fp16 = add(x = var_7519_cast_fp16, y = var_7525_cast_fp16)[name = string("value_39_cast_fp16")]; - tensor var_7528_to_fp16 = const()[name = string("op_7528_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255627520)))]; - tensor query_79_cast_fp16 = add(x = query_77_cast_fp16, y = var_7528_to_fp16)[name = string("query_79_cast_fp16")]; - tensor var_7531_to_fp16 = const()[name = string("op_7531_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255629632)))]; - tensor q_with_bias_v_39_cast_fp16 = add(x = query_77_cast_fp16, y = var_7531_to_fp16)[name = string("q_with_bias_v_39_cast_fp16")]; - string var_7541_pad_type_0 = const()[name = string("op_7541_pad_type_0"), val = string("valid")]; - tensor var_7541_strides_0 = const()[name = string("op_7541_strides_0"), val = tensor([1, 1])]; - tensor var_7541_pad_0 = const()[name = string("op_7541_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7541_dilations_0 = const()[name = string("op_7541_dilations_0"), val = tensor([1, 1])]; - int32 var_7541_groups_0 = const()[name = string("op_7541_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255631744))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256025024))))[name = string("layers_19_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_7541_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7541_dilations_0, groups = var_7541_groups_0, pad = var_7541_pad_0, pad_type = var_7541_pad_type_0, strides = var_7541_strides_0, weight = layers_19_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_7541_cast_fp16")]; - string var_7547_pad_type_0 = const()[name = string("op_7547_pad_type_0"), val = string("valid")]; - tensor var_7547_strides_0 = const()[name = string("op_7547_strides_0"), val = tensor([1, 1])]; - tensor var_7547_pad_0 = const()[name = string("op_7547_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7547_dilations_0 = const()[name = string("op_7547_dilations_0"), val = tensor([1, 1])]; - int32 var_7547_groups_0 = const()[name = string("op_7547_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256050368))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256026112))))[name = string("layers_19_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7547_cast_fp16 = conv(dilations = var_7547_dilations_0, groups = var_7547_groups_0, pad = var_7547_pad_0, pad_type = var_7547_pad_type_0, strides = var_7547_strides_0, weight = layers_19_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_7547_cast_fp16")]; - tensor p_39_cast_fp16 = add(x = var_7541_cast_fp16, y = var_7547_cast_fp16)[name = string("p_39_cast_fp16")]; - tensor var_7551 = const()[name = string("op_7551"), val = tensor([1, 8, 128, 188])]; - tensor var_7552_cast_fp16 = reshape(shape = var_7551, x = q_with_bias_v_39_cast_fp16)[name = string("op_7552_cast_fp16")]; - tensor var_7553 = const()[name = string("op_7553"), val = tensor([1, 8, 128, -1])]; - tensor var_7554_cast_fp16 = reshape(shape = var_7553, x = p_39_cast_fp16)[name = string("op_7554_cast_fp16")]; - bool matrix_bd_153_transpose_x_0 = const()[name = string("matrix_bd_153_transpose_x_0"), val = bool(true)]; - bool matrix_bd_153_transpose_y_0 = const()[name = string("matrix_bd_153_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_153_cast_fp16 = matmul(transpose_x = matrix_bd_153_transpose_x_0, transpose_y = matrix_bd_153_transpose_y_0, x = var_7552_cast_fp16, y = var_7554_cast_fp16)[name = string("matrix_bd_153_cast_fp16")]; - tensor matrix_bd_155_pad_0 = const()[name = string("matrix_bd_155_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_155_mode_0 = const()[name = string("matrix_bd_155_mode_0"), val = string("constant")]; - fp16 const_219_to_fp16 = const()[name = string("const_219_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_155_cast_fp16 = pad(constant_val = const_219_to_fp16, mode = matrix_bd_155_mode_0, pad = matrix_bd_155_pad_0, x = matrix_bd_153_cast_fp16)[name = string("matrix_bd_155_cast_fp16")]; - tensor var_7563 = const()[name = string("op_7563"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_157_cast_fp16 = reshape(shape = var_7563, x = matrix_bd_155_cast_fp16)[name = string("matrix_bd_157_cast_fp16")]; - tensor var_7567_begin_0 = const()[name = string("op_7567_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_7567_end_0 = const()[name = string("op_7567_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_7567_end_mask_0 = const()[name = string("op_7567_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_7567_cast_fp16 = slice_by_index(begin = var_7567_begin_0, end = var_7567_end_0, end_mask = var_7567_end_mask_0, x = matrix_bd_157_cast_fp16)[name = string("op_7567_cast_fp16")]; - tensor var_7568 = const()[name = string("op_7568"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_159_cast_fp16 = reshape(shape = var_7568, x = var_7567_cast_fp16)[name = string("matrix_bd_159_cast_fp16")]; - tensor var_7573_begin_0 = const()[name = string("op_7573_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7573_end_0 = const()[name = string("op_7573_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_7573_end_mask_0 = const()[name = string("op_7573_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_7573_cast_fp16 = slice_by_index(begin = var_7573_begin_0, end = var_7573_end_0, end_mask = var_7573_end_mask_0, x = matrix_bd_159_cast_fp16)[name = string("op_7573_cast_fp16")]; - fp16 var_7574_to_fp16 = const()[name = string("op_7574_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_39_cast_fp16 = mul(x = var_7573_cast_fp16, y = var_7574_to_fp16)[name = string("qk_mask_39_cast_fp16")]; - tensor var_7578 = const()[name = string("op_7578"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_39_cast_fp16 = reshape(shape = var_7578, x = query_79_cast_fp16)[name = string("mh_q_39_cast_fp16")]; - fp16 var_7580_to_fp16 = const()[name = string("op_7580_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_7581_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_7580_to_fp16)[name = string("op_7581_cast_fp16")]; - tensor var_7584 = const()[name = string("op_7584"), val = tensor([1, 8, 128, 188])]; - tensor var_7585_cast_fp16 = reshape(shape = var_7584, x = key_39_cast_fp16)[name = string("op_7585_cast_fp16")]; - bool mh_w_77_transpose_x_0 = const()[name = string("mh_w_77_transpose_x_0"), val = bool(true)]; - bool mh_w_77_transpose_y_0 = const()[name = string("mh_w_77_transpose_y_0"), val = bool(false)]; - tensor mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_7581_cast_fp16, y = var_7585_cast_fp16)[name = string("mh_w_77_cast_fp16")]; - tensor mh_w_79_cast_fp16 = add(x = mh_w_77_cast_fp16, y = qk_mask_39_cast_fp16)[name = string("mh_w_79_cast_fp16")]; - tensor var_7589_cast_fp16 = softmax(axis = var_7376, x = mh_w_79_cast_fp16)[name = string("op_7589_cast_fp16")]; - tensor var_7590 = const()[name = string("op_7590"), val = tensor([1, 8, 128, 188])]; - tensor var_7591_cast_fp16 = reshape(shape = var_7590, x = value_39_cast_fp16)[name = string("op_7591_cast_fp16")]; - bool attn_39_transpose_x_0 = const()[name = string("attn_39_transpose_x_0"), val = bool(false)]; - bool attn_39_transpose_y_0 = const()[name = string("attn_39_transpose_y_0"), val = bool(true)]; - tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_7591_cast_fp16, y = var_7589_cast_fp16)[name = string("attn_39_cast_fp16")]; - tensor var_7594 = const()[name = string("op_7594"), val = tensor([1, 1024, 1, 188])]; - tensor input_517_cast_fp16 = reshape(shape = var_7594, x = attn_39_cast_fp16)[name = string("input_517_cast_fp16")]; - string var_7604_pad_type_0 = const()[name = string("op_7604_pad_type_0"), val = string("valid")]; - tensor var_7604_strides_0 = const()[name = string("op_7604_strides_0"), val = tensor([1, 1])]; - tensor var_7604_pad_0 = const()[name = string("op_7604_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7604_dilations_0 = const()[name = string("op_7604_dilations_0"), val = tensor([1, 1])]; - int32 var_7604_groups_0 = const()[name = string("op_7604_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256181504))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256574784))))[name = string("layers_19_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7604_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7604_dilations_0, groups = var_7604_groups_0, pad = var_7604_pad_0, pad_type = var_7604_pad_type_0, strides = var_7604_strides_0, weight = layers_19_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_517_cast_fp16)[name = string("op_7604_cast_fp16")]; - string var_7610_pad_type_0 = const()[name = string("op_7610_pad_type_0"), val = string("valid")]; - tensor var_7610_strides_0 = const()[name = string("op_7610_strides_0"), val = tensor([1, 1])]; - tensor var_7610_pad_0 = const()[name = string("op_7610_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7610_dilations_0 = const()[name = string("op_7610_dilations_0"), val = tensor([1, 1])]; - int32 var_7610_groups_0 = const()[name = string("op_7610_groups_0"), val = int32(1)]; - tensor layers_19_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256583936))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256575872))))[name = string("layers_19_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7610_cast_fp16 = conv(dilations = var_7610_dilations_0, groups = var_7610_groups_0, pad = var_7610_pad_0, pad_type = var_7610_pad_type_0, strides = var_7610_strides_0, weight = layers_19_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_517_cast_fp16)[name = string("op_7610_cast_fp16")]; - tensor obj_81_cast_fp16 = add(x = var_7604_cast_fp16, y = var_7610_cast_fp16)[name = string("obj_81_cast_fp16")]; - tensor inputs_195_cast_fp16 = add(x = inputs_193_cast_fp16, y = obj_81_cast_fp16)[name = string("inputs_195_cast_fp16")]; - tensor out_195_axes_0 = const()[name = string("out_195_axes_0"), val = tensor([1])]; - fp16 var_7621_to_fp16 = const()[name = string("op_7621_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_195_cast_fp16 = layer_norm(axes = out_195_axes_0, epsilon = var_7621_to_fp16, x = inputs_195_cast_fp16)[name = string("out_195_cast_fp16")]; - tensor input_519_gamma_0_to_fp16 = const()[name = string("input_519_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256715072)))]; - tensor input_519_beta_0_to_fp16 = const()[name = string("input_519_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256717184)))]; - fp16 input_519_epsilon_0_to_fp16 = const()[name = string("input_519_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_519_cast_fp16 = batch_norm(beta = input_519_beta_0_to_fp16, epsilon = input_519_epsilon_0_to_fp16, gamma = input_519_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_195_cast_fp16)[name = string("input_519_cast_fp16")]; - string var_7642_pad_type_0 = const()[name = string("op_7642_pad_type_0"), val = string("valid")]; - tensor var_7642_strides_0 = const()[name = string("op_7642_strides_0"), val = tensor([1, 1])]; - tensor var_7642_pad_0 = const()[name = string("op_7642_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7642_dilations_0 = const()[name = string("op_7642_dilations_0"), val = tensor([1, 1])]; - int32 var_7642_groups_0 = const()[name = string("op_7642_groups_0"), val = int32(1)]; - tensor layers_19_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256719296))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257505792))))[name = string("layers_19_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_7642_cast_fp16 = conv(dilations = var_7642_dilations_0, groups = var_7642_groups_0, pad = var_7642_pad_0, pad_type = var_7642_pad_type_0, strides = var_7642_strides_0, weight = layers_19_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_519_cast_fp16)[name = string("op_7642_cast_fp16")]; - string var_7648_pad_type_0 = const()[name = string("op_7648_pad_type_0"), val = string("valid")]; - tensor var_7648_strides_0 = const()[name = string("op_7648_strides_0"), val = tensor([1, 1])]; - tensor var_7648_pad_0 = const()[name = string("op_7648_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7648_dilations_0 = const()[name = string("op_7648_dilations_0"), val = tensor([1, 1])]; - int32 var_7648_groups_0 = const()[name = string("op_7648_groups_0"), val = int32(1)]; - tensor layers_19_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257528832))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257507904))))[name = string("layers_19_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7648_cast_fp16 = conv(dilations = var_7648_dilations_0, groups = var_7648_groups_0, pad = var_7648_pad_0, pad_type = var_7648_pad_type_0, strides = var_7648_strides_0, weight = layers_19_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_519_cast_fp16)[name = string("op_7648_cast_fp16")]; - tensor input_521_cast_fp16 = add(x = var_7642_cast_fp16, y = var_7648_cast_fp16)[name = string("input_521_cast_fp16")]; - int32 input_523_split_num_splits_0 = const()[name = string("input_523_split_num_splits_0"), val = int32(2)]; - int32 input_523_split_axis_0 = const()[name = string("input_523_split_axis_0"), val = int32(1)]; - tensor input_523_split_cast_fp16_0, tensor input_523_split_cast_fp16_1 = split(axis = input_523_split_axis_0, num_splits = input_523_split_num_splits_0, x = input_521_cast_fp16)[name = string("input_523_split_cast_fp16")]; - tensor input_523_split_1_sigmoid_cast_fp16 = sigmoid(x = input_523_split_cast_fp16_1)[name = string("input_523_split_1_sigmoid_cast_fp16")]; - tensor input_523_cast_fp16 = mul(x = input_523_split_cast_fp16_0, y = input_523_split_1_sigmoid_cast_fp16)[name = string("input_523_cast_fp16")]; - string input_525_pad_type_0 = const()[name = string("input_525_pad_type_0"), val = string("custom")]; - tensor input_525_pad_0 = const()[name = string("input_525_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_525_groups_0 = const()[name = string("input_525_groups_0"), val = int32(1024)]; - tensor input_525_strides_0 = const()[name = string("input_525_strides_0"), val = tensor([1, 1])]; - tensor input_525_dilations_0 = const()[name = string("input_525_dilations_0"), val = tensor([1, 1])]; - tensor const_306_to_fp16 = const()[name = string("const_306_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257791040)))]; - tensor const_307_to_fp16 = const()[name = string("const_307_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257809536)))]; - tensor input_527_cast_fp16 = conv(bias = const_307_to_fp16, dilations = input_525_dilations_0, groups = input_525_groups_0, pad = input_525_pad_0, pad_type = input_525_pad_type_0, strides = input_525_strides_0, weight = const_306_to_fp16, x = input_523_cast_fp16)[name = string("input_527_cast_fp16")]; - tensor input_529_cast_fp16 = silu(x = input_527_cast_fp16)[name = string("input_529_cast_fp16")]; - string var_7670_pad_type_0 = const()[name = string("op_7670_pad_type_0"), val = string("valid")]; - tensor var_7670_strides_0 = const()[name = string("op_7670_strides_0"), val = tensor([1, 1])]; - tensor var_7670_pad_0 = const()[name = string("op_7670_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7670_dilations_0 = const()[name = string("op_7670_dilations_0"), val = tensor([1, 1])]; - int32 var_7670_groups_0 = const()[name = string("op_7670_groups_0"), val = int32(1)]; - tensor layers_19_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257811648))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258204928))))[name = string("layers_19_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_7670_cast_fp16 = conv(dilations = var_7670_dilations_0, groups = var_7670_groups_0, pad = var_7670_pad_0, pad_type = var_7670_pad_type_0, strides = var_7670_strides_0, weight = layers_19_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_529_cast_fp16)[name = string("op_7670_cast_fp16")]; - string var_7676_pad_type_0 = const()[name = string("op_7676_pad_type_0"), val = string("valid")]; - tensor var_7676_strides_0 = const()[name = string("op_7676_strides_0"), val = tensor([1, 1])]; - tensor var_7676_pad_0 = const()[name = string("op_7676_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7676_dilations_0 = const()[name = string("op_7676_dilations_0"), val = tensor([1, 1])]; - int32 var_7676_groups_0 = const()[name = string("op_7676_groups_0"), val = int32(1)]; - tensor layers_19_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258215040))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258206016))))[name = string("layers_19_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7676_cast_fp16 = conv(dilations = var_7676_dilations_0, groups = var_7676_groups_0, pad = var_7676_pad_0, pad_type = var_7676_pad_type_0, strides = var_7676_strides_0, weight = layers_19_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_529_cast_fp16)[name = string("op_7676_cast_fp16")]; - tensor x_119_cast_fp16 = add(x = var_7670_cast_fp16, y = var_7676_cast_fp16)[name = string("x_119_cast_fp16")]; - tensor inputs_197_cast_fp16 = add(x = inputs_195_cast_fp16, y = x_119_cast_fp16)[name = string("inputs_197_cast_fp16")]; - tensor out_197_axes_0 = const()[name = string("out_197_axes_0"), val = tensor([1])]; - fp16 var_7687_to_fp16 = const()[name = string("op_7687_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_197_cast_fp16 = layer_norm(axes = out_197_axes_0, epsilon = var_7687_to_fp16, x = inputs_197_cast_fp16)[name = string("out_197_cast_fp16")]; - tensor input_531_gamma_0_to_fp16 = const()[name = string("input_531_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258346176)))]; - tensor input_531_beta_0_to_fp16 = const()[name = string("input_531_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258348288)))]; - fp16 input_531_epsilon_0_to_fp16 = const()[name = string("input_531_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_531_cast_fp16 = batch_norm(beta = input_531_beta_0_to_fp16, epsilon = input_531_epsilon_0_to_fp16, gamma = input_531_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_197_cast_fp16)[name = string("input_531_cast_fp16")]; - string var_7707_pad_type_0 = const()[name = string("op_7707_pad_type_0"), val = string("valid")]; - tensor var_7707_strides_0 = const()[name = string("op_7707_strides_0"), val = tensor([1, 1])]; - tensor var_7707_pad_0 = const()[name = string("op_7707_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7707_dilations_0 = const()[name = string("op_7707_dilations_0"), val = tensor([1, 1])]; - int32 var_7707_groups_0 = const()[name = string("op_7707_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258350400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259923328))))[name = string("layers_19_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_7707_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7707_dilations_0, groups = var_7707_groups_0, pad = var_7707_pad_0, pad_type = var_7707_pad_type_0, strides = var_7707_strides_0, weight = layers_19_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_531_cast_fp16)[name = string("op_7707_cast_fp16")]; - string var_7713_pad_type_0 = const()[name = string("op_7713_pad_type_0"), val = string("valid")]; - tensor var_7713_strides_0 = const()[name = string("op_7713_strides_0"), val = tensor([1, 1])]; - tensor var_7713_pad_0 = const()[name = string("op_7713_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7713_dilations_0 = const()[name = string("op_7713_dilations_0"), val = tensor([1, 1])]; - int32 var_7713_groups_0 = const()[name = string("op_7713_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259964352))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259927488))))[name = string("layers_19_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7713_cast_fp16 = conv(dilations = var_7713_dilations_0, groups = var_7713_groups_0, pad = var_7713_pad_0, pad_type = var_7713_pad_type_0, strides = var_7713_strides_0, weight = layers_19_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_531_cast_fp16)[name = string("op_7713_cast_fp16")]; - tensor input_533_cast_fp16 = add(x = var_7707_cast_fp16, y = var_7713_cast_fp16)[name = string("input_533_cast_fp16")]; - tensor input_535_cast_fp16 = silu(x = input_533_cast_fp16)[name = string("input_535_cast_fp16")]; - string var_7724_pad_type_0 = const()[name = string("op_7724_pad_type_0"), val = string("valid")]; - tensor var_7724_strides_0 = const()[name = string("op_7724_strides_0"), val = tensor([1, 1])]; - tensor var_7724_pad_0 = const()[name = string("op_7724_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7724_dilations_0 = const()[name = string("op_7724_dilations_0"), val = tensor([1, 1])]; - int32 var_7724_groups_0 = const()[name = string("op_7724_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260488704))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262061632))))[name = string("layers_19_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_7724_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7724_dilations_0, groups = var_7724_groups_0, pad = var_7724_pad_0, pad_type = var_7724_pad_type_0, strides = var_7724_strides_0, weight = layers_19_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_535_cast_fp16)[name = string("op_7724_cast_fp16")]; - string var_7730_pad_type_0 = const()[name = string("op_7730_pad_type_0"), val = string("valid")]; - tensor var_7730_strides_0 = const()[name = string("op_7730_strides_0"), val = tensor([1, 1])]; - tensor var_7730_pad_0 = const()[name = string("op_7730_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7730_dilations_0 = const()[name = string("op_7730_dilations_0"), val = tensor([1, 1])]; - int32 var_7730_groups_0 = const()[name = string("op_7730_groups_0"), val = int32(1)]; - tensor layers_19_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262105280))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262062720))))[name = string("layers_19_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7730_cast_fp16 = conv(dilations = var_7730_dilations_0, groups = var_7730_groups_0, pad = var_7730_pad_0, pad_type = var_7730_pad_type_0, strides = var_7730_strides_0, weight = layers_19_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_535_cast_fp16)[name = string("op_7730_cast_fp16")]; - tensor x_121_cast_fp16 = add(x = var_7724_cast_fp16, y = var_7730_cast_fp16)[name = string("x_121_cast_fp16")]; - fp16 var_7732_to_fp16 = const()[name = string("op_7732_to_fp16"), val = fp16(0x1p-1)]; - tensor var_7733_cast_fp16 = mul(x = x_121_cast_fp16, y = var_7732_to_fp16)[name = string("op_7733_cast_fp16")]; - tensor inputs_199_cast_fp16 = add(x = inputs_197_cast_fp16, y = var_7733_cast_fp16)[name = string("inputs_199_cast_fp16")]; - tensor out_199_axes_0 = const()[name = string("out_199_axes_0"), val = tensor([1])]; - fp16 var_7743_to_fp16 = const()[name = string("op_7743_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_199_cast_fp16 = layer_norm(axes = out_199_axes_0, epsilon = var_7743_to_fp16, x = inputs_199_cast_fp16)[name = string("out_199_cast_fp16")]; - tensor inputs_201_gamma_0_to_fp16 = const()[name = string("inputs_201_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262629632)))]; - tensor inputs_201_beta_0_to_fp16 = const()[name = string("inputs_201_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262631744)))]; - fp16 inputs_201_epsilon_0_to_fp16 = const()[name = string("inputs_201_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_201_cast_fp16 = batch_norm(beta = inputs_201_beta_0_to_fp16, epsilon = inputs_201_epsilon_0_to_fp16, gamma = inputs_201_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_199_cast_fp16)[name = string("inputs_201_cast_fp16")]; - int32 var_7757 = const()[name = string("op_7757"), val = int32(3)]; - tensor out_201_axes_0 = const()[name = string("out_201_axes_0"), val = tensor([1])]; - fp16 var_7788_to_fp16 = const()[name = string("op_7788_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_201_cast_fp16 = layer_norm(axes = out_201_axes_0, epsilon = var_7788_to_fp16, x = inputs_201_cast_fp16)[name = string("out_201_cast_fp16")]; - tensor input_537_gamma_0_to_fp16 = const()[name = string("input_537_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262633856)))]; - tensor input_537_beta_0_to_fp16 = const()[name = string("input_537_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262635968)))]; - fp16 input_537_epsilon_0_to_fp16 = const()[name = string("input_537_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_537_cast_fp16 = batch_norm(beta = input_537_beta_0_to_fp16, epsilon = input_537_epsilon_0_to_fp16, gamma = input_537_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_201_cast_fp16)[name = string("input_537_cast_fp16")]; - string var_7808_pad_type_0 = const()[name = string("op_7808_pad_type_0"), val = string("valid")]; - tensor var_7808_strides_0 = const()[name = string("op_7808_strides_0"), val = tensor([1, 1])]; - tensor var_7808_pad_0 = const()[name = string("op_7808_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7808_dilations_0 = const()[name = string("op_7808_dilations_0"), val = tensor([1, 1])]; - int32 var_7808_groups_0 = const()[name = string("op_7808_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262638080))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264211008))))[name = string("layers_20_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_7808_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7808_dilations_0, groups = var_7808_groups_0, pad = var_7808_pad_0, pad_type = var_7808_pad_type_0, strides = var_7808_strides_0, weight = layers_20_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_537_cast_fp16)[name = string("op_7808_cast_fp16")]; - string var_7814_pad_type_0 = const()[name = string("op_7814_pad_type_0"), val = string("valid")]; - tensor var_7814_strides_0 = const()[name = string("op_7814_strides_0"), val = tensor([1, 1])]; - tensor var_7814_pad_0 = const()[name = string("op_7814_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7814_dilations_0 = const()[name = string("op_7814_dilations_0"), val = tensor([1, 1])]; - int32 var_7814_groups_0 = const()[name = string("op_7814_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264252736))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264215168))))[name = string("layers_20_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7814_cast_fp16 = conv(dilations = var_7814_dilations_0, groups = var_7814_groups_0, pad = var_7814_pad_0, pad_type = var_7814_pad_type_0, strides = var_7814_strides_0, weight = layers_20_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_537_cast_fp16)[name = string("op_7814_cast_fp16")]; - tensor input_539_cast_fp16 = add(x = var_7808_cast_fp16, y = var_7814_cast_fp16)[name = string("input_539_cast_fp16")]; - tensor input_541_cast_fp16 = silu(x = input_539_cast_fp16)[name = string("input_541_cast_fp16")]; - string var_7825_pad_type_0 = const()[name = string("op_7825_pad_type_0"), val = string("valid")]; - tensor var_7825_strides_0 = const()[name = string("op_7825_strides_0"), val = tensor([1, 1])]; - tensor var_7825_pad_0 = const()[name = string("op_7825_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7825_dilations_0 = const()[name = string("op_7825_dilations_0"), val = tensor([1, 1])]; - int32 var_7825_groups_0 = const()[name = string("op_7825_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264777088))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266350016))))[name = string("layers_20_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_7825_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7825_dilations_0, groups = var_7825_groups_0, pad = var_7825_pad_0, pad_type = var_7825_pad_type_0, strides = var_7825_strides_0, weight = layers_20_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_541_cast_fp16)[name = string("op_7825_cast_fp16")]; - string var_7831_pad_type_0 = const()[name = string("op_7831_pad_type_0"), val = string("valid")]; - tensor var_7831_strides_0 = const()[name = string("op_7831_strides_0"), val = tensor([1, 1])]; - tensor var_7831_pad_0 = const()[name = string("op_7831_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7831_dilations_0 = const()[name = string("op_7831_dilations_0"), val = tensor([1, 1])]; - int32 var_7831_groups_0 = const()[name = string("op_7831_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266397760))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266351104))))[name = string("layers_20_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7831_cast_fp16 = conv(dilations = var_7831_dilations_0, groups = var_7831_groups_0, pad = var_7831_pad_0, pad_type = var_7831_pad_type_0, strides = var_7831_strides_0, weight = layers_20_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_541_cast_fp16)[name = string("op_7831_cast_fp16")]; - tensor x_123_cast_fp16 = add(x = var_7825_cast_fp16, y = var_7831_cast_fp16)[name = string("x_123_cast_fp16")]; - fp16 var_7833_to_fp16 = const()[name = string("op_7833_to_fp16"), val = fp16(0x1p-1)]; - tensor var_7834_cast_fp16 = mul(x = x_123_cast_fp16, y = var_7833_to_fp16)[name = string("op_7834_cast_fp16")]; - tensor inputs_203_cast_fp16 = add(x = inputs_201_cast_fp16, y = var_7834_cast_fp16)[name = string("inputs_203_cast_fp16")]; - tensor out_203_axes_0 = const()[name = string("out_203_axes_0"), val = tensor([1])]; - fp16 var_7844_to_fp16 = const()[name = string("op_7844_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_203_cast_fp16 = layer_norm(axes = out_203_axes_0, epsilon = var_7844_to_fp16, x = inputs_203_cast_fp16)[name = string("out_203_cast_fp16")]; - tensor obj_83_gamma_0_to_fp16 = const()[name = string("obj_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266922112)))]; - tensor obj_83_beta_0_to_fp16 = const()[name = string("obj_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266924224)))]; - fp16 obj_83_epsilon_0_to_fp16 = const()[name = string("obj_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_83_cast_fp16 = batch_norm(beta = obj_83_beta_0_to_fp16, epsilon = obj_83_epsilon_0_to_fp16, gamma = obj_83_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_203_cast_fp16)[name = string("obj_83_cast_fp16")]; - string var_7869_pad_type_0 = const()[name = string("op_7869_pad_type_0"), val = string("valid")]; - tensor var_7869_strides_0 = const()[name = string("op_7869_strides_0"), val = tensor([1, 1])]; - tensor var_7869_pad_0 = const()[name = string("op_7869_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7869_dilations_0 = const()[name = string("op_7869_dilations_0"), val = tensor([1, 1])]; - int32 var_7869_groups_0 = const()[name = string("op_7869_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266926336))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267319616))))[name = string("layers_20_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7869_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7869_dilations_0, groups = var_7869_groups_0, pad = var_7869_pad_0, pad_type = var_7869_pad_type_0, strides = var_7869_strides_0, weight = layers_20_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_83_cast_fp16)[name = string("op_7869_cast_fp16")]; - string var_7875_pad_type_0 = const()[name = string("op_7875_pad_type_0"), val = string("valid")]; - tensor var_7875_strides_0 = const()[name = string("op_7875_strides_0"), val = tensor([1, 1])]; - tensor var_7875_pad_0 = const()[name = string("op_7875_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7875_dilations_0 = const()[name = string("op_7875_dilations_0"), val = tensor([1, 1])]; - int32 var_7875_groups_0 = const()[name = string("op_7875_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267329152))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267320704))))[name = string("layers_20_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7875_cast_fp16 = conv(dilations = var_7875_dilations_0, groups = var_7875_groups_0, pad = var_7875_pad_0, pad_type = var_7875_pad_type_0, strides = var_7875_strides_0, weight = layers_20_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_83_cast_fp16)[name = string("op_7875_cast_fp16")]; - tensor query_81_cast_fp16 = add(x = var_7869_cast_fp16, y = var_7875_cast_fp16)[name = string("query_81_cast_fp16")]; - string var_7884_pad_type_0 = const()[name = string("op_7884_pad_type_0"), val = string("valid")]; - tensor var_7884_strides_0 = const()[name = string("op_7884_strides_0"), val = tensor([1, 1])]; - tensor var_7884_pad_0 = const()[name = string("op_7884_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7884_dilations_0 = const()[name = string("op_7884_dilations_0"), val = tensor([1, 1])]; - int32 var_7884_groups_0 = const()[name = string("op_7884_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267460288))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267853568))))[name = string("layers_20_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7884_cast_fp16 = conv(dilations = var_7884_dilations_0, groups = var_7884_groups_0, pad = var_7884_pad_0, pad_type = var_7884_pad_type_0, strides = var_7884_strides_0, weight = layers_20_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_83_cast_fp16)[name = string("op_7884_cast_fp16")]; - string var_7890_pad_type_0 = const()[name = string("op_7890_pad_type_0"), val = string("valid")]; - tensor var_7890_strides_0 = const()[name = string("op_7890_strides_0"), val = tensor([1, 1])]; - tensor var_7890_pad_0 = const()[name = string("op_7890_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7890_dilations_0 = const()[name = string("op_7890_dilations_0"), val = tensor([1, 1])]; - int32 var_7890_groups_0 = const()[name = string("op_7890_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267865216))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267854656))))[name = string("layers_20_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7890_cast_fp16 = conv(dilations = var_7890_dilations_0, groups = var_7890_groups_0, pad = var_7890_pad_0, pad_type = var_7890_pad_type_0, strides = var_7890_strides_0, weight = layers_20_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_83_cast_fp16)[name = string("op_7890_cast_fp16")]; - tensor key_41_cast_fp16 = add(x = var_7884_cast_fp16, y = var_7890_cast_fp16)[name = string("key_41_cast_fp16")]; - string var_7900_pad_type_0 = const()[name = string("op_7900_pad_type_0"), val = string("valid")]; - tensor var_7900_strides_0 = const()[name = string("op_7900_strides_0"), val = tensor([1, 1])]; - tensor var_7900_pad_0 = const()[name = string("op_7900_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7900_dilations_0 = const()[name = string("op_7900_dilations_0"), val = tensor([1, 1])]; - int32 var_7900_groups_0 = const()[name = string("op_7900_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267996352))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268389632))))[name = string("layers_20_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7900_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7900_dilations_0, groups = var_7900_groups_0, pad = var_7900_pad_0, pad_type = var_7900_pad_type_0, strides = var_7900_strides_0, weight = layers_20_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_83_cast_fp16)[name = string("op_7900_cast_fp16")]; - string var_7906_pad_type_0 = const()[name = string("op_7906_pad_type_0"), val = string("valid")]; - tensor var_7906_strides_0 = const()[name = string("op_7906_strides_0"), val = tensor([1, 1])]; - tensor var_7906_pad_0 = const()[name = string("op_7906_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7906_dilations_0 = const()[name = string("op_7906_dilations_0"), val = tensor([1, 1])]; - int32 var_7906_groups_0 = const()[name = string("op_7906_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268397632))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268390720))))[name = string("layers_20_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7906_cast_fp16 = conv(dilations = var_7906_dilations_0, groups = var_7906_groups_0, pad = var_7906_pad_0, pad_type = var_7906_pad_type_0, strides = var_7906_strides_0, weight = layers_20_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_83_cast_fp16)[name = string("op_7906_cast_fp16")]; - tensor value_41_cast_fp16 = add(x = var_7900_cast_fp16, y = var_7906_cast_fp16)[name = string("value_41_cast_fp16")]; - tensor var_7909_to_fp16 = const()[name = string("op_7909_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268528768)))]; - tensor query_83_cast_fp16 = add(x = query_81_cast_fp16, y = var_7909_to_fp16)[name = string("query_83_cast_fp16")]; - tensor var_7912_to_fp16 = const()[name = string("op_7912_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268530880)))]; - tensor q_with_bias_v_41_cast_fp16 = add(x = query_81_cast_fp16, y = var_7912_to_fp16)[name = string("q_with_bias_v_41_cast_fp16")]; - string var_7922_pad_type_0 = const()[name = string("op_7922_pad_type_0"), val = string("valid")]; - tensor var_7922_strides_0 = const()[name = string("op_7922_strides_0"), val = tensor([1, 1])]; - tensor var_7922_pad_0 = const()[name = string("op_7922_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7922_dilations_0 = const()[name = string("op_7922_dilations_0"), val = tensor([1, 1])]; - int32 var_7922_groups_0 = const()[name = string("op_7922_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268532992))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268926272))))[name = string("layers_20_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_7922_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7922_dilations_0, groups = var_7922_groups_0, pad = var_7922_pad_0, pad_type = var_7922_pad_type_0, strides = var_7922_strides_0, weight = layers_20_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_7922_cast_fp16")]; - string var_7928_pad_type_0 = const()[name = string("op_7928_pad_type_0"), val = string("valid")]; - tensor var_7928_strides_0 = const()[name = string("op_7928_strides_0"), val = tensor([1, 1])]; - tensor var_7928_pad_0 = const()[name = string("op_7928_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7928_dilations_0 = const()[name = string("op_7928_dilations_0"), val = tensor([1, 1])]; - int32 var_7928_groups_0 = const()[name = string("op_7928_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268955264))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268927360))))[name = string("layers_20_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7928_cast_fp16 = conv(dilations = var_7928_dilations_0, groups = var_7928_groups_0, pad = var_7928_pad_0, pad_type = var_7928_pad_type_0, strides = var_7928_strides_0, weight = layers_20_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_7928_cast_fp16")]; - tensor p_41_cast_fp16 = add(x = var_7922_cast_fp16, y = var_7928_cast_fp16)[name = string("p_41_cast_fp16")]; - tensor var_7932 = const()[name = string("op_7932"), val = tensor([1, 8, 128, 188])]; - tensor var_7933_cast_fp16 = reshape(shape = var_7932, x = q_with_bias_v_41_cast_fp16)[name = string("op_7933_cast_fp16")]; - tensor var_7934 = const()[name = string("op_7934"), val = tensor([1, 8, 128, -1])]; - tensor var_7935_cast_fp16 = reshape(shape = var_7934, x = p_41_cast_fp16)[name = string("op_7935_cast_fp16")]; - bool matrix_bd_161_transpose_x_0 = const()[name = string("matrix_bd_161_transpose_x_0"), val = bool(true)]; - bool matrix_bd_161_transpose_y_0 = const()[name = string("matrix_bd_161_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_161_cast_fp16 = matmul(transpose_x = matrix_bd_161_transpose_x_0, transpose_y = matrix_bd_161_transpose_y_0, x = var_7933_cast_fp16, y = var_7935_cast_fp16)[name = string("matrix_bd_161_cast_fp16")]; - tensor matrix_bd_163_pad_0 = const()[name = string("matrix_bd_163_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_163_mode_0 = const()[name = string("matrix_bd_163_mode_0"), val = string("constant")]; - fp16 const_230_to_fp16 = const()[name = string("const_230_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_163_cast_fp16 = pad(constant_val = const_230_to_fp16, mode = matrix_bd_163_mode_0, pad = matrix_bd_163_pad_0, x = matrix_bd_161_cast_fp16)[name = string("matrix_bd_163_cast_fp16")]; - tensor var_7944 = const()[name = string("op_7944"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_165_cast_fp16 = reshape(shape = var_7944, x = matrix_bd_163_cast_fp16)[name = string("matrix_bd_165_cast_fp16")]; - tensor var_7948_begin_0 = const()[name = string("op_7948_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_7948_end_0 = const()[name = string("op_7948_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_7948_end_mask_0 = const()[name = string("op_7948_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_7948_cast_fp16 = slice_by_index(begin = var_7948_begin_0, end = var_7948_end_0, end_mask = var_7948_end_mask_0, x = matrix_bd_165_cast_fp16)[name = string("op_7948_cast_fp16")]; - tensor var_7949 = const()[name = string("op_7949"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_167_cast_fp16 = reshape(shape = var_7949, x = var_7948_cast_fp16)[name = string("matrix_bd_167_cast_fp16")]; - tensor var_7954_begin_0 = const()[name = string("op_7954_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7954_end_0 = const()[name = string("op_7954_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_7954_end_mask_0 = const()[name = string("op_7954_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_7954_cast_fp16 = slice_by_index(begin = var_7954_begin_0, end = var_7954_end_0, end_mask = var_7954_end_mask_0, x = matrix_bd_167_cast_fp16)[name = string("op_7954_cast_fp16")]; - fp16 var_7955_to_fp16 = const()[name = string("op_7955_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_41_cast_fp16 = mul(x = var_7954_cast_fp16, y = var_7955_to_fp16)[name = string("qk_mask_41_cast_fp16")]; - tensor var_7959 = const()[name = string("op_7959"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_41_cast_fp16 = reshape(shape = var_7959, x = query_83_cast_fp16)[name = string("mh_q_41_cast_fp16")]; - fp16 var_7961_to_fp16 = const()[name = string("op_7961_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_7962_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_7961_to_fp16)[name = string("op_7962_cast_fp16")]; - tensor var_7965 = const()[name = string("op_7965"), val = tensor([1, 8, 128, 188])]; - tensor var_7966_cast_fp16 = reshape(shape = var_7965, x = key_41_cast_fp16)[name = string("op_7966_cast_fp16")]; - bool mh_w_81_transpose_x_0 = const()[name = string("mh_w_81_transpose_x_0"), val = bool(true)]; - bool mh_w_81_transpose_y_0 = const()[name = string("mh_w_81_transpose_y_0"), val = bool(false)]; - tensor mh_w_81_cast_fp16 = matmul(transpose_x = mh_w_81_transpose_x_0, transpose_y = mh_w_81_transpose_y_0, x = var_7962_cast_fp16, y = var_7966_cast_fp16)[name = string("mh_w_81_cast_fp16")]; - tensor mh_w_83_cast_fp16 = add(x = mh_w_81_cast_fp16, y = qk_mask_41_cast_fp16)[name = string("mh_w_83_cast_fp16")]; - tensor var_7970_cast_fp16 = softmax(axis = var_7757, x = mh_w_83_cast_fp16)[name = string("op_7970_cast_fp16")]; - tensor var_7971 = const()[name = string("op_7971"), val = tensor([1, 8, 128, 188])]; - tensor var_7972_cast_fp16 = reshape(shape = var_7971, x = value_41_cast_fp16)[name = string("op_7972_cast_fp16")]; - bool attn_41_transpose_x_0 = const()[name = string("attn_41_transpose_x_0"), val = bool(false)]; - bool attn_41_transpose_y_0 = const()[name = string("attn_41_transpose_y_0"), val = bool(true)]; - tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_7972_cast_fp16, y = var_7970_cast_fp16)[name = string("attn_41_cast_fp16")]; - tensor var_7975 = const()[name = string("op_7975"), val = tensor([1, 1024, 1, 188])]; - tensor input_543_cast_fp16 = reshape(shape = var_7975, x = attn_41_cast_fp16)[name = string("input_543_cast_fp16")]; - string var_7985_pad_type_0 = const()[name = string("op_7985_pad_type_0"), val = string("valid")]; - tensor var_7985_strides_0 = const()[name = string("op_7985_strides_0"), val = tensor([1, 1])]; - tensor var_7985_pad_0 = const()[name = string("op_7985_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7985_dilations_0 = const()[name = string("op_7985_dilations_0"), val = tensor([1, 1])]; - int32 var_7985_groups_0 = const()[name = string("op_7985_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269086400))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269479680))))[name = string("layers_20_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_7985_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7985_dilations_0, groups = var_7985_groups_0, pad = var_7985_pad_0, pad_type = var_7985_pad_type_0, strides = var_7985_strides_0, weight = layers_20_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_543_cast_fp16)[name = string("op_7985_cast_fp16")]; - string var_7991_pad_type_0 = const()[name = string("op_7991_pad_type_0"), val = string("valid")]; - tensor var_7991_strides_0 = const()[name = string("op_7991_strides_0"), val = tensor([1, 1])]; - tensor var_7991_pad_0 = const()[name = string("op_7991_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_7991_dilations_0 = const()[name = string("op_7991_dilations_0"), val = tensor([1, 1])]; - int32 var_7991_groups_0 = const()[name = string("op_7991_groups_0"), val = int32(1)]; - tensor layers_20_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269489088))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269480768))))[name = string("layers_20_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_7991_cast_fp16 = conv(dilations = var_7991_dilations_0, groups = var_7991_groups_0, pad = var_7991_pad_0, pad_type = var_7991_pad_type_0, strides = var_7991_strides_0, weight = layers_20_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_543_cast_fp16)[name = string("op_7991_cast_fp16")]; - tensor obj_85_cast_fp16 = add(x = var_7985_cast_fp16, y = var_7991_cast_fp16)[name = string("obj_85_cast_fp16")]; - tensor inputs_205_cast_fp16 = add(x = inputs_203_cast_fp16, y = obj_85_cast_fp16)[name = string("inputs_205_cast_fp16")]; - tensor out_205_axes_0 = const()[name = string("out_205_axes_0"), val = tensor([1])]; - fp16 var_8002_to_fp16 = const()[name = string("op_8002_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_205_cast_fp16 = layer_norm(axes = out_205_axes_0, epsilon = var_8002_to_fp16, x = inputs_205_cast_fp16)[name = string("out_205_cast_fp16")]; - tensor input_545_gamma_0_to_fp16 = const()[name = string("input_545_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269620224)))]; - tensor input_545_beta_0_to_fp16 = const()[name = string("input_545_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269622336)))]; - fp16 input_545_epsilon_0_to_fp16 = const()[name = string("input_545_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_545_cast_fp16 = batch_norm(beta = input_545_beta_0_to_fp16, epsilon = input_545_epsilon_0_to_fp16, gamma = input_545_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_205_cast_fp16)[name = string("input_545_cast_fp16")]; - string var_8023_pad_type_0 = const()[name = string("op_8023_pad_type_0"), val = string("valid")]; - tensor var_8023_strides_0 = const()[name = string("op_8023_strides_0"), val = tensor([1, 1])]; - tensor var_8023_pad_0 = const()[name = string("op_8023_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8023_dilations_0 = const()[name = string("op_8023_dilations_0"), val = tensor([1, 1])]; - int32 var_8023_groups_0 = const()[name = string("op_8023_groups_0"), val = int32(1)]; - tensor layers_20_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269624448))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270410944))))[name = string("layers_20_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8023_cast_fp16 = conv(dilations = var_8023_dilations_0, groups = var_8023_groups_0, pad = var_8023_pad_0, pad_type = var_8023_pad_type_0, strides = var_8023_strides_0, weight = layers_20_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_545_cast_fp16)[name = string("op_8023_cast_fp16")]; - string var_8029_pad_type_0 = const()[name = string("op_8029_pad_type_0"), val = string("valid")]; - tensor var_8029_strides_0 = const()[name = string("op_8029_strides_0"), val = tensor([1, 1])]; - tensor var_8029_pad_0 = const()[name = string("op_8029_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8029_dilations_0 = const()[name = string("op_8029_dilations_0"), val = tensor([1, 1])]; - int32 var_8029_groups_0 = const()[name = string("op_8029_groups_0"), val = int32(1)]; - tensor layers_20_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270432576))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270413056))))[name = string("layers_20_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8029_cast_fp16 = conv(dilations = var_8029_dilations_0, groups = var_8029_groups_0, pad = var_8029_pad_0, pad_type = var_8029_pad_type_0, strides = var_8029_strides_0, weight = layers_20_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_545_cast_fp16)[name = string("op_8029_cast_fp16")]; - tensor input_547_cast_fp16 = add(x = var_8023_cast_fp16, y = var_8029_cast_fp16)[name = string("input_547_cast_fp16")]; - int32 input_549_split_num_splits_0 = const()[name = string("input_549_split_num_splits_0"), val = int32(2)]; - int32 input_549_split_axis_0 = const()[name = string("input_549_split_axis_0"), val = int32(1)]; - tensor input_549_split_cast_fp16_0, tensor input_549_split_cast_fp16_1 = split(axis = input_549_split_axis_0, num_splits = input_549_split_num_splits_0, x = input_547_cast_fp16)[name = string("input_549_split_cast_fp16")]; - tensor input_549_split_1_sigmoid_cast_fp16 = sigmoid(x = input_549_split_cast_fp16_1)[name = string("input_549_split_1_sigmoid_cast_fp16")]; - tensor input_549_cast_fp16 = mul(x = input_549_split_cast_fp16_0, y = input_549_split_1_sigmoid_cast_fp16)[name = string("input_549_cast_fp16")]; - string input_551_pad_type_0 = const()[name = string("input_551_pad_type_0"), val = string("custom")]; - tensor input_551_pad_0 = const()[name = string("input_551_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_551_groups_0 = const()[name = string("input_551_groups_0"), val = int32(1024)]; - tensor input_551_strides_0 = const()[name = string("input_551_strides_0"), val = tensor([1, 1])]; - tensor input_551_dilations_0 = const()[name = string("input_551_dilations_0"), val = tensor([1, 1])]; - tensor const_308_to_fp16 = const()[name = string("const_308_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270694784)))]; - tensor const_309_to_fp16 = const()[name = string("const_309_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270713280)))]; - tensor input_553_cast_fp16 = conv(bias = const_309_to_fp16, dilations = input_551_dilations_0, groups = input_551_groups_0, pad = input_551_pad_0, pad_type = input_551_pad_type_0, strides = input_551_strides_0, weight = const_308_to_fp16, x = input_549_cast_fp16)[name = string("input_553_cast_fp16")]; - tensor input_555_cast_fp16 = silu(x = input_553_cast_fp16)[name = string("input_555_cast_fp16")]; - string var_8051_pad_type_0 = const()[name = string("op_8051_pad_type_0"), val = string("valid")]; - tensor var_8051_strides_0 = const()[name = string("op_8051_strides_0"), val = tensor([1, 1])]; - tensor var_8051_pad_0 = const()[name = string("op_8051_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8051_dilations_0 = const()[name = string("op_8051_dilations_0"), val = tensor([1, 1])]; - int32 var_8051_groups_0 = const()[name = string("op_8051_groups_0"), val = int32(1)]; - tensor layers_20_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270715392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271108672))))[name = string("layers_20_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8051_cast_fp16 = conv(dilations = var_8051_dilations_0, groups = var_8051_groups_0, pad = var_8051_pad_0, pad_type = var_8051_pad_type_0, strides = var_8051_strides_0, weight = layers_20_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_555_cast_fp16)[name = string("op_8051_cast_fp16")]; - string var_8057_pad_type_0 = const()[name = string("op_8057_pad_type_0"), val = string("valid")]; - tensor var_8057_strides_0 = const()[name = string("op_8057_strides_0"), val = tensor([1, 1])]; - tensor var_8057_pad_0 = const()[name = string("op_8057_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8057_dilations_0 = const()[name = string("op_8057_dilations_0"), val = tensor([1, 1])]; - int32 var_8057_groups_0 = const()[name = string("op_8057_groups_0"), val = int32(1)]; - tensor layers_20_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271119552))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271109760))))[name = string("layers_20_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8057_cast_fp16 = conv(dilations = var_8057_dilations_0, groups = var_8057_groups_0, pad = var_8057_pad_0, pad_type = var_8057_pad_type_0, strides = var_8057_strides_0, weight = layers_20_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_555_cast_fp16)[name = string("op_8057_cast_fp16")]; - tensor x_125_cast_fp16 = add(x = var_8051_cast_fp16, y = var_8057_cast_fp16)[name = string("x_125_cast_fp16")]; - tensor inputs_207_cast_fp16 = add(x = inputs_205_cast_fp16, y = x_125_cast_fp16)[name = string("inputs_207_cast_fp16")]; - tensor out_207_axes_0 = const()[name = string("out_207_axes_0"), val = tensor([1])]; - fp16 var_8068_to_fp16 = const()[name = string("op_8068_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_207_cast_fp16 = layer_norm(axes = out_207_axes_0, epsilon = var_8068_to_fp16, x = inputs_207_cast_fp16)[name = string("out_207_cast_fp16")]; - tensor input_557_gamma_0_to_fp16 = const()[name = string("input_557_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271250688)))]; - tensor input_557_beta_0_to_fp16 = const()[name = string("input_557_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271252800)))]; - fp16 input_557_epsilon_0_to_fp16 = const()[name = string("input_557_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_557_cast_fp16 = batch_norm(beta = input_557_beta_0_to_fp16, epsilon = input_557_epsilon_0_to_fp16, gamma = input_557_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_207_cast_fp16)[name = string("input_557_cast_fp16")]; - string var_8088_pad_type_0 = const()[name = string("op_8088_pad_type_0"), val = string("valid")]; - tensor var_8088_strides_0 = const()[name = string("op_8088_strides_0"), val = tensor([1, 1])]; - tensor var_8088_pad_0 = const()[name = string("op_8088_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8088_dilations_0 = const()[name = string("op_8088_dilations_0"), val = tensor([1, 1])]; - int32 var_8088_groups_0 = const()[name = string("op_8088_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271254912))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272827840))))[name = string("layers_20_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8088_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8088_dilations_0, groups = var_8088_groups_0, pad = var_8088_pad_0, pad_type = var_8088_pad_type_0, strides = var_8088_strides_0, weight = layers_20_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_557_cast_fp16)[name = string("op_8088_cast_fp16")]; - string var_8094_pad_type_0 = const()[name = string("op_8094_pad_type_0"), val = string("valid")]; - tensor var_8094_strides_0 = const()[name = string("op_8094_strides_0"), val = tensor([1, 1])]; - tensor var_8094_pad_0 = const()[name = string("op_8094_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8094_dilations_0 = const()[name = string("op_8094_dilations_0"), val = tensor([1, 1])]; - int32 var_8094_groups_0 = const()[name = string("op_8094_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272867264))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272832000))))[name = string("layers_20_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8094_cast_fp16 = conv(dilations = var_8094_dilations_0, groups = var_8094_groups_0, pad = var_8094_pad_0, pad_type = var_8094_pad_type_0, strides = var_8094_strides_0, weight = layers_20_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_557_cast_fp16)[name = string("op_8094_cast_fp16")]; - tensor input_559_cast_fp16 = add(x = var_8088_cast_fp16, y = var_8094_cast_fp16)[name = string("input_559_cast_fp16")]; - tensor input_561_cast_fp16 = silu(x = input_559_cast_fp16)[name = string("input_561_cast_fp16")]; - string var_8105_pad_type_0 = const()[name = string("op_8105_pad_type_0"), val = string("valid")]; - tensor var_8105_strides_0 = const()[name = string("op_8105_strides_0"), val = tensor([1, 1])]; - tensor var_8105_pad_0 = const()[name = string("op_8105_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8105_dilations_0 = const()[name = string("op_8105_dilations_0"), val = tensor([1, 1])]; - int32 var_8105_groups_0 = const()[name = string("op_8105_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273391616))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274964544))))[name = string("layers_20_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8105_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8105_dilations_0, groups = var_8105_groups_0, pad = var_8105_pad_0, pad_type = var_8105_pad_type_0, strides = var_8105_strides_0, weight = layers_20_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_561_cast_fp16)[name = string("op_8105_cast_fp16")]; - string var_8111_pad_type_0 = const()[name = string("op_8111_pad_type_0"), val = string("valid")]; - tensor var_8111_strides_0 = const()[name = string("op_8111_strides_0"), val = tensor([1, 1])]; - tensor var_8111_pad_0 = const()[name = string("op_8111_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8111_dilations_0 = const()[name = string("op_8111_dilations_0"), val = tensor([1, 1])]; - int32 var_8111_groups_0 = const()[name = string("op_8111_groups_0"), val = int32(1)]; - tensor layers_20_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275010560))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274965632))))[name = string("layers_20_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8111_cast_fp16 = conv(dilations = var_8111_dilations_0, groups = var_8111_groups_0, pad = var_8111_pad_0, pad_type = var_8111_pad_type_0, strides = var_8111_strides_0, weight = layers_20_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_561_cast_fp16)[name = string("op_8111_cast_fp16")]; - tensor x_127_cast_fp16 = add(x = var_8105_cast_fp16, y = var_8111_cast_fp16)[name = string("x_127_cast_fp16")]; - fp16 var_8113_to_fp16 = const()[name = string("op_8113_to_fp16"), val = fp16(0x1p-1)]; - tensor var_8114_cast_fp16 = mul(x = x_127_cast_fp16, y = var_8113_to_fp16)[name = string("op_8114_cast_fp16")]; - tensor inputs_209_cast_fp16 = add(x = inputs_207_cast_fp16, y = var_8114_cast_fp16)[name = string("inputs_209_cast_fp16")]; - tensor out_209_axes_0 = const()[name = string("out_209_axes_0"), val = tensor([1])]; - fp16 var_8124_to_fp16 = const()[name = string("op_8124_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_209_cast_fp16 = layer_norm(axes = out_209_axes_0, epsilon = var_8124_to_fp16, x = inputs_209_cast_fp16)[name = string("out_209_cast_fp16")]; - tensor inputs_211_gamma_0_to_fp16 = const()[name = string("inputs_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275534912)))]; - tensor inputs_211_beta_0_to_fp16 = const()[name = string("inputs_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275537024)))]; - fp16 inputs_211_epsilon_0_to_fp16 = const()[name = string("inputs_211_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_211_cast_fp16 = batch_norm(beta = inputs_211_beta_0_to_fp16, epsilon = inputs_211_epsilon_0_to_fp16, gamma = inputs_211_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_209_cast_fp16)[name = string("inputs_211_cast_fp16")]; - int32 var_8138 = const()[name = string("op_8138"), val = int32(3)]; - tensor out_211_axes_0 = const()[name = string("out_211_axes_0"), val = tensor([1])]; - fp16 var_8169_to_fp16 = const()[name = string("op_8169_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_211_cast_fp16 = layer_norm(axes = out_211_axes_0, epsilon = var_8169_to_fp16, x = inputs_211_cast_fp16)[name = string("out_211_cast_fp16")]; - tensor input_563_gamma_0_to_fp16 = const()[name = string("input_563_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275539136)))]; - tensor input_563_beta_0_to_fp16 = const()[name = string("input_563_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275541248)))]; - fp16 input_563_epsilon_0_to_fp16 = const()[name = string("input_563_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_563_cast_fp16 = batch_norm(beta = input_563_beta_0_to_fp16, epsilon = input_563_epsilon_0_to_fp16, gamma = input_563_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_211_cast_fp16)[name = string("input_563_cast_fp16")]; - string var_8189_pad_type_0 = const()[name = string("op_8189_pad_type_0"), val = string("valid")]; - tensor var_8189_strides_0 = const()[name = string("op_8189_strides_0"), val = tensor([1, 1])]; - tensor var_8189_pad_0 = const()[name = string("op_8189_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8189_dilations_0 = const()[name = string("op_8189_dilations_0"), val = tensor([1, 1])]; - int32 var_8189_groups_0 = const()[name = string("op_8189_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275543360))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277116288))))[name = string("layers_21_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8189_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8189_dilations_0, groups = var_8189_groups_0, pad = var_8189_pad_0, pad_type = var_8189_pad_type_0, strides = var_8189_strides_0, weight = layers_21_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_563_cast_fp16)[name = string("op_8189_cast_fp16")]; - string var_8195_pad_type_0 = const()[name = string("op_8195_pad_type_0"), val = string("valid")]; - tensor var_8195_strides_0 = const()[name = string("op_8195_strides_0"), val = tensor([1, 1])]; - tensor var_8195_pad_0 = const()[name = string("op_8195_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8195_dilations_0 = const()[name = string("op_8195_dilations_0"), val = tensor([1, 1])]; - int32 var_8195_groups_0 = const()[name = string("op_8195_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277158144))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277120448))))[name = string("layers_21_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8195_cast_fp16 = conv(dilations = var_8195_dilations_0, groups = var_8195_groups_0, pad = var_8195_pad_0, pad_type = var_8195_pad_type_0, strides = var_8195_strides_0, weight = layers_21_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_563_cast_fp16)[name = string("op_8195_cast_fp16")]; - tensor input_565_cast_fp16 = add(x = var_8189_cast_fp16, y = var_8195_cast_fp16)[name = string("input_565_cast_fp16")]; - tensor input_567_cast_fp16 = silu(x = input_565_cast_fp16)[name = string("input_567_cast_fp16")]; - string var_8206_pad_type_0 = const()[name = string("op_8206_pad_type_0"), val = string("valid")]; - tensor var_8206_strides_0 = const()[name = string("op_8206_strides_0"), val = tensor([1, 1])]; - tensor var_8206_pad_0 = const()[name = string("op_8206_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8206_dilations_0 = const()[name = string("op_8206_dilations_0"), val = tensor([1, 1])]; - int32 var_8206_groups_0 = const()[name = string("op_8206_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277682496))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279255424))))[name = string("layers_21_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8206_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8206_dilations_0, groups = var_8206_groups_0, pad = var_8206_pad_0, pad_type = var_8206_pad_type_0, strides = var_8206_strides_0, weight = layers_21_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_567_cast_fp16)[name = string("op_8206_cast_fp16")]; - string var_8212_pad_type_0 = const()[name = string("op_8212_pad_type_0"), val = string("valid")]; - tensor var_8212_strides_0 = const()[name = string("op_8212_strides_0"), val = tensor([1, 1])]; - tensor var_8212_pad_0 = const()[name = string("op_8212_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8212_dilations_0 = const()[name = string("op_8212_dilations_0"), val = tensor([1, 1])]; - int32 var_8212_groups_0 = const()[name = string("op_8212_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279306560))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279256512))))[name = string("layers_21_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8212_cast_fp16 = conv(dilations = var_8212_dilations_0, groups = var_8212_groups_0, pad = var_8212_pad_0, pad_type = var_8212_pad_type_0, strides = var_8212_strides_0, weight = layers_21_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_567_cast_fp16)[name = string("op_8212_cast_fp16")]; - tensor x_129_cast_fp16 = add(x = var_8206_cast_fp16, y = var_8212_cast_fp16)[name = string("x_129_cast_fp16")]; - fp16 var_8214_to_fp16 = const()[name = string("op_8214_to_fp16"), val = fp16(0x1p-1)]; - tensor var_8215_cast_fp16 = mul(x = x_129_cast_fp16, y = var_8214_to_fp16)[name = string("op_8215_cast_fp16")]; - tensor inputs_213_cast_fp16 = add(x = inputs_211_cast_fp16, y = var_8215_cast_fp16)[name = string("inputs_213_cast_fp16")]; - tensor out_213_axes_0 = const()[name = string("out_213_axes_0"), val = tensor([1])]; - fp16 var_8225_to_fp16 = const()[name = string("op_8225_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_213_cast_fp16 = layer_norm(axes = out_213_axes_0, epsilon = var_8225_to_fp16, x = inputs_213_cast_fp16)[name = string("out_213_cast_fp16")]; - tensor obj_87_gamma_0_to_fp16 = const()[name = string("obj_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279830912)))]; - tensor obj_87_beta_0_to_fp16 = const()[name = string("obj_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279833024)))]; - fp16 obj_87_epsilon_0_to_fp16 = const()[name = string("obj_87_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_87_cast_fp16 = batch_norm(beta = obj_87_beta_0_to_fp16, epsilon = obj_87_epsilon_0_to_fp16, gamma = obj_87_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_213_cast_fp16)[name = string("obj_87_cast_fp16")]; - string var_8250_pad_type_0 = const()[name = string("op_8250_pad_type_0"), val = string("valid")]; - tensor var_8250_strides_0 = const()[name = string("op_8250_strides_0"), val = tensor([1, 1])]; - tensor var_8250_pad_0 = const()[name = string("op_8250_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8250_dilations_0 = const()[name = string("op_8250_dilations_0"), val = tensor([1, 1])]; - int32 var_8250_groups_0 = const()[name = string("op_8250_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279835136))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280228416))))[name = string("layers_21_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8250_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8250_dilations_0, groups = var_8250_groups_0, pad = var_8250_pad_0, pad_type = var_8250_pad_type_0, strides = var_8250_strides_0, weight = layers_21_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_87_cast_fp16)[name = string("op_8250_cast_fp16")]; - string var_8256_pad_type_0 = const()[name = string("op_8256_pad_type_0"), val = string("valid")]; - tensor var_8256_strides_0 = const()[name = string("op_8256_strides_0"), val = tensor([1, 1])]; - tensor var_8256_pad_0 = const()[name = string("op_8256_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8256_dilations_0 = const()[name = string("op_8256_dilations_0"), val = tensor([1, 1])]; - int32 var_8256_groups_0 = const()[name = string("op_8256_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280239808))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280229504))))[name = string("layers_21_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8256_cast_fp16 = conv(dilations = var_8256_dilations_0, groups = var_8256_groups_0, pad = var_8256_pad_0, pad_type = var_8256_pad_type_0, strides = var_8256_strides_0, weight = layers_21_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_87_cast_fp16)[name = string("op_8256_cast_fp16")]; - tensor query_85_cast_fp16 = add(x = var_8250_cast_fp16, y = var_8256_cast_fp16)[name = string("query_85_cast_fp16")]; - string var_8265_pad_type_0 = const()[name = string("op_8265_pad_type_0"), val = string("valid")]; - tensor var_8265_strides_0 = const()[name = string("op_8265_strides_0"), val = tensor([1, 1])]; - tensor var_8265_pad_0 = const()[name = string("op_8265_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8265_dilations_0 = const()[name = string("op_8265_dilations_0"), val = tensor([1, 1])]; - int32 var_8265_groups_0 = const()[name = string("op_8265_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280370944))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280764224))))[name = string("layers_21_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8265_cast_fp16 = conv(dilations = var_8265_dilations_0, groups = var_8265_groups_0, pad = var_8265_pad_0, pad_type = var_8265_pad_type_0, strides = var_8265_strides_0, weight = layers_21_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_87_cast_fp16)[name = string("op_8265_cast_fp16")]; - string var_8271_pad_type_0 = const()[name = string("op_8271_pad_type_0"), val = string("valid")]; - tensor var_8271_strides_0 = const()[name = string("op_8271_strides_0"), val = tensor([1, 1])]; - tensor var_8271_pad_0 = const()[name = string("op_8271_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8271_dilations_0 = const()[name = string("op_8271_dilations_0"), val = tensor([1, 1])]; - int32 var_8271_groups_0 = const()[name = string("op_8271_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280777152))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280765312))))[name = string("layers_21_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8271_cast_fp16 = conv(dilations = var_8271_dilations_0, groups = var_8271_groups_0, pad = var_8271_pad_0, pad_type = var_8271_pad_type_0, strides = var_8271_strides_0, weight = layers_21_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_87_cast_fp16)[name = string("op_8271_cast_fp16")]; - tensor key_43_cast_fp16 = add(x = var_8265_cast_fp16, y = var_8271_cast_fp16)[name = string("key_43_cast_fp16")]; - string var_8281_pad_type_0 = const()[name = string("op_8281_pad_type_0"), val = string("valid")]; - tensor var_8281_strides_0 = const()[name = string("op_8281_strides_0"), val = tensor([1, 1])]; - tensor var_8281_pad_0 = const()[name = string("op_8281_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8281_dilations_0 = const()[name = string("op_8281_dilations_0"), val = tensor([1, 1])]; - int32 var_8281_groups_0 = const()[name = string("op_8281_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280908288))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281301568))))[name = string("layers_21_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8281_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8281_dilations_0, groups = var_8281_groups_0, pad = var_8281_pad_0, pad_type = var_8281_pad_type_0, strides = var_8281_strides_0, weight = layers_21_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_87_cast_fp16)[name = string("op_8281_cast_fp16")]; - string var_8287_pad_type_0 = const()[name = string("op_8287_pad_type_0"), val = string("valid")]; - tensor var_8287_strides_0 = const()[name = string("op_8287_strides_0"), val = tensor([1, 1])]; - tensor var_8287_pad_0 = const()[name = string("op_8287_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8287_dilations_0 = const()[name = string("op_8287_dilations_0"), val = tensor([1, 1])]; - int32 var_8287_groups_0 = const()[name = string("op_8287_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281310656))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281302656))))[name = string("layers_21_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8287_cast_fp16 = conv(dilations = var_8287_dilations_0, groups = var_8287_groups_0, pad = var_8287_pad_0, pad_type = var_8287_pad_type_0, strides = var_8287_strides_0, weight = layers_21_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_87_cast_fp16)[name = string("op_8287_cast_fp16")]; - tensor value_43_cast_fp16 = add(x = var_8281_cast_fp16, y = var_8287_cast_fp16)[name = string("value_43_cast_fp16")]; - tensor var_8290_to_fp16 = const()[name = string("op_8290_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281441792)))]; - tensor query_87_cast_fp16 = add(x = query_85_cast_fp16, y = var_8290_to_fp16)[name = string("query_87_cast_fp16")]; - tensor var_8293_to_fp16 = const()[name = string("op_8293_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281443904)))]; - tensor q_with_bias_v_43_cast_fp16 = add(x = query_85_cast_fp16, y = var_8293_to_fp16)[name = string("q_with_bias_v_43_cast_fp16")]; - string var_8303_pad_type_0 = const()[name = string("op_8303_pad_type_0"), val = string("valid")]; - tensor var_8303_strides_0 = const()[name = string("op_8303_strides_0"), val = tensor([1, 1])]; - tensor var_8303_pad_0 = const()[name = string("op_8303_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8303_dilations_0 = const()[name = string("op_8303_dilations_0"), val = tensor([1, 1])]; - int32 var_8303_groups_0 = const()[name = string("op_8303_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281446016))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281839296))))[name = string("layers_21_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_8303_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8303_dilations_0, groups = var_8303_groups_0, pad = var_8303_pad_0, pad_type = var_8303_pad_type_0, strides = var_8303_strides_0, weight = layers_21_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_8303_cast_fp16")]; - string var_8309_pad_type_0 = const()[name = string("op_8309_pad_type_0"), val = string("valid")]; - tensor var_8309_strides_0 = const()[name = string("op_8309_strides_0"), val = tensor([1, 1])]; - tensor var_8309_pad_0 = const()[name = string("op_8309_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8309_dilations_0 = const()[name = string("op_8309_dilations_0"), val = tensor([1, 1])]; - int32 var_8309_groups_0 = const()[name = string("op_8309_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281866304))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281840384))))[name = string("layers_21_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8309_cast_fp16 = conv(dilations = var_8309_dilations_0, groups = var_8309_groups_0, pad = var_8309_pad_0, pad_type = var_8309_pad_type_0, strides = var_8309_strides_0, weight = layers_21_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_8309_cast_fp16")]; - tensor p_43_cast_fp16 = add(x = var_8303_cast_fp16, y = var_8309_cast_fp16)[name = string("p_43_cast_fp16")]; - tensor var_8313 = const()[name = string("op_8313"), val = tensor([1, 8, 128, 188])]; - tensor var_8314_cast_fp16 = reshape(shape = var_8313, x = q_with_bias_v_43_cast_fp16)[name = string("op_8314_cast_fp16")]; - tensor var_8315 = const()[name = string("op_8315"), val = tensor([1, 8, 128, -1])]; - tensor var_8316_cast_fp16 = reshape(shape = var_8315, x = p_43_cast_fp16)[name = string("op_8316_cast_fp16")]; - bool matrix_bd_169_transpose_x_0 = const()[name = string("matrix_bd_169_transpose_x_0"), val = bool(true)]; - bool matrix_bd_169_transpose_y_0 = const()[name = string("matrix_bd_169_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_169_cast_fp16 = matmul(transpose_x = matrix_bd_169_transpose_x_0, transpose_y = matrix_bd_169_transpose_y_0, x = var_8314_cast_fp16, y = var_8316_cast_fp16)[name = string("matrix_bd_169_cast_fp16")]; - tensor matrix_bd_171_pad_0 = const()[name = string("matrix_bd_171_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_171_mode_0 = const()[name = string("matrix_bd_171_mode_0"), val = string("constant")]; - fp16 const_241_to_fp16 = const()[name = string("const_241_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_171_cast_fp16 = pad(constant_val = const_241_to_fp16, mode = matrix_bd_171_mode_0, pad = matrix_bd_171_pad_0, x = matrix_bd_169_cast_fp16)[name = string("matrix_bd_171_cast_fp16")]; - tensor var_8325 = const()[name = string("op_8325"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_173_cast_fp16 = reshape(shape = var_8325, x = matrix_bd_171_cast_fp16)[name = string("matrix_bd_173_cast_fp16")]; - tensor var_8329_begin_0 = const()[name = string("op_8329_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_8329_end_0 = const()[name = string("op_8329_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_8329_end_mask_0 = const()[name = string("op_8329_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_8329_cast_fp16 = slice_by_index(begin = var_8329_begin_0, end = var_8329_end_0, end_mask = var_8329_end_mask_0, x = matrix_bd_173_cast_fp16)[name = string("op_8329_cast_fp16")]; - tensor var_8330 = const()[name = string("op_8330"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_175_cast_fp16 = reshape(shape = var_8330, x = var_8329_cast_fp16)[name = string("matrix_bd_175_cast_fp16")]; - tensor var_8335_begin_0 = const()[name = string("op_8335_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8335_end_0 = const()[name = string("op_8335_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_8335_end_mask_0 = const()[name = string("op_8335_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_8335_cast_fp16 = slice_by_index(begin = var_8335_begin_0, end = var_8335_end_0, end_mask = var_8335_end_mask_0, x = matrix_bd_175_cast_fp16)[name = string("op_8335_cast_fp16")]; - fp16 var_8336_to_fp16 = const()[name = string("op_8336_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_43_cast_fp16 = mul(x = var_8335_cast_fp16, y = var_8336_to_fp16)[name = string("qk_mask_43_cast_fp16")]; - tensor var_8340 = const()[name = string("op_8340"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_43_cast_fp16 = reshape(shape = var_8340, x = query_87_cast_fp16)[name = string("mh_q_43_cast_fp16")]; - fp16 var_8342_to_fp16 = const()[name = string("op_8342_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_8343_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_8342_to_fp16)[name = string("op_8343_cast_fp16")]; - tensor var_8346 = const()[name = string("op_8346"), val = tensor([1, 8, 128, 188])]; - tensor var_8347_cast_fp16 = reshape(shape = var_8346, x = key_43_cast_fp16)[name = string("op_8347_cast_fp16")]; - bool mh_w_85_transpose_x_0 = const()[name = string("mh_w_85_transpose_x_0"), val = bool(true)]; - bool mh_w_85_transpose_y_0 = const()[name = string("mh_w_85_transpose_y_0"), val = bool(false)]; - tensor mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_8343_cast_fp16, y = var_8347_cast_fp16)[name = string("mh_w_85_cast_fp16")]; - tensor mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = qk_mask_43_cast_fp16)[name = string("mh_w_87_cast_fp16")]; - tensor var_8351_cast_fp16 = softmax(axis = var_8138, x = mh_w_87_cast_fp16)[name = string("op_8351_cast_fp16")]; - tensor var_8352 = const()[name = string("op_8352"), val = tensor([1, 8, 128, 188])]; - tensor var_8353_cast_fp16 = reshape(shape = var_8352, x = value_43_cast_fp16)[name = string("op_8353_cast_fp16")]; - bool attn_43_transpose_x_0 = const()[name = string("attn_43_transpose_x_0"), val = bool(false)]; - bool attn_43_transpose_y_0 = const()[name = string("attn_43_transpose_y_0"), val = bool(true)]; - tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_8353_cast_fp16, y = var_8351_cast_fp16)[name = string("attn_43_cast_fp16")]; - tensor var_8356 = const()[name = string("op_8356"), val = tensor([1, 1024, 1, 188])]; - tensor input_569_cast_fp16 = reshape(shape = var_8356, x = attn_43_cast_fp16)[name = string("input_569_cast_fp16")]; - string var_8366_pad_type_0 = const()[name = string("op_8366_pad_type_0"), val = string("valid")]; - tensor var_8366_strides_0 = const()[name = string("op_8366_strides_0"), val = tensor([1, 1])]; - tensor var_8366_pad_0 = const()[name = string("op_8366_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8366_dilations_0 = const()[name = string("op_8366_dilations_0"), val = tensor([1, 1])]; - int32 var_8366_groups_0 = const()[name = string("op_8366_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281997440))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282390720))))[name = string("layers_21_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8366_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8366_dilations_0, groups = var_8366_groups_0, pad = var_8366_pad_0, pad_type = var_8366_pad_type_0, strides = var_8366_strides_0, weight = layers_21_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_569_cast_fp16)[name = string("op_8366_cast_fp16")]; - string var_8372_pad_type_0 = const()[name = string("op_8372_pad_type_0"), val = string("valid")]; - tensor var_8372_strides_0 = const()[name = string("op_8372_strides_0"), val = tensor([1, 1])]; - tensor var_8372_pad_0 = const()[name = string("op_8372_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8372_dilations_0 = const()[name = string("op_8372_dilations_0"), val = tensor([1, 1])]; - int32 var_8372_groups_0 = const()[name = string("op_8372_groups_0"), val = int32(1)]; - tensor layers_21_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282400640))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282391808))))[name = string("layers_21_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8372_cast_fp16 = conv(dilations = var_8372_dilations_0, groups = var_8372_groups_0, pad = var_8372_pad_0, pad_type = var_8372_pad_type_0, strides = var_8372_strides_0, weight = layers_21_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_569_cast_fp16)[name = string("op_8372_cast_fp16")]; - tensor obj_89_cast_fp16 = add(x = var_8366_cast_fp16, y = var_8372_cast_fp16)[name = string("obj_89_cast_fp16")]; - tensor inputs_215_cast_fp16 = add(x = inputs_213_cast_fp16, y = obj_89_cast_fp16)[name = string("inputs_215_cast_fp16")]; - tensor out_215_axes_0 = const()[name = string("out_215_axes_0"), val = tensor([1])]; - fp16 var_8383_to_fp16 = const()[name = string("op_8383_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_215_cast_fp16 = layer_norm(axes = out_215_axes_0, epsilon = var_8383_to_fp16, x = inputs_215_cast_fp16)[name = string("out_215_cast_fp16")]; - tensor input_571_gamma_0_to_fp16 = const()[name = string("input_571_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282531776)))]; - tensor input_571_beta_0_to_fp16 = const()[name = string("input_571_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282533888)))]; - fp16 input_571_epsilon_0_to_fp16 = const()[name = string("input_571_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_571_cast_fp16 = batch_norm(beta = input_571_beta_0_to_fp16, epsilon = input_571_epsilon_0_to_fp16, gamma = input_571_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_215_cast_fp16)[name = string("input_571_cast_fp16")]; - string var_8404_pad_type_0 = const()[name = string("op_8404_pad_type_0"), val = string("valid")]; - tensor var_8404_strides_0 = const()[name = string("op_8404_strides_0"), val = tensor([1, 1])]; - tensor var_8404_pad_0 = const()[name = string("op_8404_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8404_dilations_0 = const()[name = string("op_8404_dilations_0"), val = tensor([1, 1])]; - int32 var_8404_groups_0 = const()[name = string("op_8404_groups_0"), val = int32(1)]; - tensor layers_21_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282536000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283322496))))[name = string("layers_21_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8404_cast_fp16 = conv(dilations = var_8404_dilations_0, groups = var_8404_groups_0, pad = var_8404_pad_0, pad_type = var_8404_pad_type_0, strides = var_8404_strides_0, weight = layers_21_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_571_cast_fp16)[name = string("op_8404_cast_fp16")]; - string var_8410_pad_type_0 = const()[name = string("op_8410_pad_type_0"), val = string("valid")]; - tensor var_8410_strides_0 = const()[name = string("op_8410_strides_0"), val = tensor([1, 1])]; - tensor var_8410_pad_0 = const()[name = string("op_8410_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8410_dilations_0 = const()[name = string("op_8410_dilations_0"), val = tensor([1, 1])]; - int32 var_8410_groups_0 = const()[name = string("op_8410_groups_0"), val = int32(1)]; - tensor layers_21_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283344128))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283324608))))[name = string("layers_21_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8410_cast_fp16 = conv(dilations = var_8410_dilations_0, groups = var_8410_groups_0, pad = var_8410_pad_0, pad_type = var_8410_pad_type_0, strides = var_8410_strides_0, weight = layers_21_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_571_cast_fp16)[name = string("op_8410_cast_fp16")]; - tensor input_573_cast_fp16 = add(x = var_8404_cast_fp16, y = var_8410_cast_fp16)[name = string("input_573_cast_fp16")]; - int32 input_575_split_num_splits_0 = const()[name = string("input_575_split_num_splits_0"), val = int32(2)]; - int32 input_575_split_axis_0 = const()[name = string("input_575_split_axis_0"), val = int32(1)]; - tensor input_575_split_cast_fp16_0, tensor input_575_split_cast_fp16_1 = split(axis = input_575_split_axis_0, num_splits = input_575_split_num_splits_0, x = input_573_cast_fp16)[name = string("input_575_split_cast_fp16")]; - tensor input_575_split_1_sigmoid_cast_fp16 = sigmoid(x = input_575_split_cast_fp16_1)[name = string("input_575_split_1_sigmoid_cast_fp16")]; - tensor input_575_cast_fp16 = mul(x = input_575_split_cast_fp16_0, y = input_575_split_1_sigmoid_cast_fp16)[name = string("input_575_cast_fp16")]; - string input_577_pad_type_0 = const()[name = string("input_577_pad_type_0"), val = string("custom")]; - tensor input_577_pad_0 = const()[name = string("input_577_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_577_groups_0 = const()[name = string("input_577_groups_0"), val = int32(1024)]; - tensor input_577_strides_0 = const()[name = string("input_577_strides_0"), val = tensor([1, 1])]; - tensor input_577_dilations_0 = const()[name = string("input_577_dilations_0"), val = tensor([1, 1])]; - tensor const_310_to_fp16 = const()[name = string("const_310_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283606336)))]; - tensor const_311_to_fp16 = const()[name = string("const_311_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283624832)))]; - tensor input_579_cast_fp16 = conv(bias = const_311_to_fp16, dilations = input_577_dilations_0, groups = input_577_groups_0, pad = input_577_pad_0, pad_type = input_577_pad_type_0, strides = input_577_strides_0, weight = const_310_to_fp16, x = input_575_cast_fp16)[name = string("input_579_cast_fp16")]; - tensor input_581_cast_fp16 = silu(x = input_579_cast_fp16)[name = string("input_581_cast_fp16")]; - string var_8432_pad_type_0 = const()[name = string("op_8432_pad_type_0"), val = string("valid")]; - tensor var_8432_strides_0 = const()[name = string("op_8432_strides_0"), val = tensor([1, 1])]; - tensor var_8432_pad_0 = const()[name = string("op_8432_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8432_dilations_0 = const()[name = string("op_8432_dilations_0"), val = tensor([1, 1])]; - int32 var_8432_groups_0 = const()[name = string("op_8432_groups_0"), val = int32(1)]; - tensor layers_21_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283626944))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284020224))))[name = string("layers_21_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8432_cast_fp16 = conv(dilations = var_8432_dilations_0, groups = var_8432_groups_0, pad = var_8432_pad_0, pad_type = var_8432_pad_type_0, strides = var_8432_strides_0, weight = layers_21_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_581_cast_fp16)[name = string("op_8432_cast_fp16")]; - string var_8438_pad_type_0 = const()[name = string("op_8438_pad_type_0"), val = string("valid")]; - tensor var_8438_strides_0 = const()[name = string("op_8438_strides_0"), val = tensor([1, 1])]; - tensor var_8438_pad_0 = const()[name = string("op_8438_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8438_dilations_0 = const()[name = string("op_8438_dilations_0"), val = tensor([1, 1])]; - int32 var_8438_groups_0 = const()[name = string("op_8438_groups_0"), val = int32(1)]; - tensor layers_21_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284032064))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284021312))))[name = string("layers_21_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8438_cast_fp16 = conv(dilations = var_8438_dilations_0, groups = var_8438_groups_0, pad = var_8438_pad_0, pad_type = var_8438_pad_type_0, strides = var_8438_strides_0, weight = layers_21_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_581_cast_fp16)[name = string("op_8438_cast_fp16")]; - tensor x_131_cast_fp16 = add(x = var_8432_cast_fp16, y = var_8438_cast_fp16)[name = string("x_131_cast_fp16")]; - tensor inputs_217_cast_fp16 = add(x = inputs_215_cast_fp16, y = x_131_cast_fp16)[name = string("inputs_217_cast_fp16")]; - tensor out_217_axes_0 = const()[name = string("out_217_axes_0"), val = tensor([1])]; - fp16 var_8449_to_fp16 = const()[name = string("op_8449_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_217_cast_fp16 = layer_norm(axes = out_217_axes_0, epsilon = var_8449_to_fp16, x = inputs_217_cast_fp16)[name = string("out_217_cast_fp16")]; - tensor input_583_gamma_0_to_fp16 = const()[name = string("input_583_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284163200)))]; - tensor input_583_beta_0_to_fp16 = const()[name = string("input_583_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284165312)))]; - fp16 input_583_epsilon_0_to_fp16 = const()[name = string("input_583_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_583_cast_fp16 = batch_norm(beta = input_583_beta_0_to_fp16, epsilon = input_583_epsilon_0_to_fp16, gamma = input_583_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_217_cast_fp16)[name = string("input_583_cast_fp16")]; - string var_8469_pad_type_0 = const()[name = string("op_8469_pad_type_0"), val = string("valid")]; - tensor var_8469_strides_0 = const()[name = string("op_8469_strides_0"), val = tensor([1, 1])]; - tensor var_8469_pad_0 = const()[name = string("op_8469_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8469_dilations_0 = const()[name = string("op_8469_dilations_0"), val = tensor([1, 1])]; - int32 var_8469_groups_0 = const()[name = string("op_8469_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284167424))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285740352))))[name = string("layers_21_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8469_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8469_dilations_0, groups = var_8469_groups_0, pad = var_8469_pad_0, pad_type = var_8469_pad_type_0, strides = var_8469_strides_0, weight = layers_21_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_583_cast_fp16)[name = string("op_8469_cast_fp16")]; - string var_8475_pad_type_0 = const()[name = string("op_8475_pad_type_0"), val = string("valid")]; - tensor var_8475_strides_0 = const()[name = string("op_8475_strides_0"), val = tensor([1, 1])]; - tensor var_8475_pad_0 = const()[name = string("op_8475_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8475_dilations_0 = const()[name = string("op_8475_dilations_0"), val = tensor([1, 1])]; - int32 var_8475_groups_0 = const()[name = string("op_8475_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285781696))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285744512))))[name = string("layers_21_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8475_cast_fp16 = conv(dilations = var_8475_dilations_0, groups = var_8475_groups_0, pad = var_8475_pad_0, pad_type = var_8475_pad_type_0, strides = var_8475_strides_0, weight = layers_21_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_583_cast_fp16)[name = string("op_8475_cast_fp16")]; - tensor input_585_cast_fp16 = add(x = var_8469_cast_fp16, y = var_8475_cast_fp16)[name = string("input_585_cast_fp16")]; - tensor input_587_cast_fp16 = silu(x = input_585_cast_fp16)[name = string("input_587_cast_fp16")]; - string var_8486_pad_type_0 = const()[name = string("op_8486_pad_type_0"), val = string("valid")]; - tensor var_8486_strides_0 = const()[name = string("op_8486_strides_0"), val = tensor([1, 1])]; - tensor var_8486_pad_0 = const()[name = string("op_8486_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8486_dilations_0 = const()[name = string("op_8486_dilations_0"), val = tensor([1, 1])]; - int32 var_8486_groups_0 = const()[name = string("op_8486_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286306048))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287878976))))[name = string("layers_21_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8486_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8486_dilations_0, groups = var_8486_groups_0, pad = var_8486_pad_0, pad_type = var_8486_pad_type_0, strides = var_8486_strides_0, weight = layers_21_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_587_cast_fp16)[name = string("op_8486_cast_fp16")]; - string var_8492_pad_type_0 = const()[name = string("op_8492_pad_type_0"), val = string("valid")]; - tensor var_8492_strides_0 = const()[name = string("op_8492_strides_0"), val = tensor([1, 1])]; - tensor var_8492_pad_0 = const()[name = string("op_8492_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8492_dilations_0 = const()[name = string("op_8492_dilations_0"), val = tensor([1, 1])]; - int32 var_8492_groups_0 = const()[name = string("op_8492_groups_0"), val = int32(1)]; - tensor layers_21_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287933504))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287880064))))[name = string("layers_21_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8492_cast_fp16 = conv(dilations = var_8492_dilations_0, groups = var_8492_groups_0, pad = var_8492_pad_0, pad_type = var_8492_pad_type_0, strides = var_8492_strides_0, weight = layers_21_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_587_cast_fp16)[name = string("op_8492_cast_fp16")]; - tensor x_133_cast_fp16 = add(x = var_8486_cast_fp16, y = var_8492_cast_fp16)[name = string("x_133_cast_fp16")]; - fp16 var_8494_to_fp16 = const()[name = string("op_8494_to_fp16"), val = fp16(0x1p-1)]; - tensor var_8495_cast_fp16 = mul(x = x_133_cast_fp16, y = var_8494_to_fp16)[name = string("op_8495_cast_fp16")]; - tensor inputs_219_cast_fp16 = add(x = inputs_217_cast_fp16, y = var_8495_cast_fp16)[name = string("inputs_219_cast_fp16")]; - tensor out_219_axes_0 = const()[name = string("out_219_axes_0"), val = tensor([1])]; - fp16 var_8505_to_fp16 = const()[name = string("op_8505_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_219_cast_fp16 = layer_norm(axes = out_219_axes_0, epsilon = var_8505_to_fp16, x = inputs_219_cast_fp16)[name = string("out_219_cast_fp16")]; - tensor inputs_221_gamma_0_to_fp16 = const()[name = string("inputs_221_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288457856)))]; - tensor inputs_221_beta_0_to_fp16 = const()[name = string("inputs_221_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288459968)))]; - fp16 inputs_221_epsilon_0_to_fp16 = const()[name = string("inputs_221_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_221_cast_fp16 = batch_norm(beta = inputs_221_beta_0_to_fp16, epsilon = inputs_221_epsilon_0_to_fp16, gamma = inputs_221_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_219_cast_fp16)[name = string("inputs_221_cast_fp16")]; - int32 var_8519 = const()[name = string("op_8519"), val = int32(3)]; - tensor out_221_axes_0 = const()[name = string("out_221_axes_0"), val = tensor([1])]; - fp16 var_8550_to_fp16 = const()[name = string("op_8550_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_221_cast_fp16 = layer_norm(axes = out_221_axes_0, epsilon = var_8550_to_fp16, x = inputs_221_cast_fp16)[name = string("out_221_cast_fp16")]; - tensor input_589_gamma_0_to_fp16 = const()[name = string("input_589_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288462080)))]; - tensor input_589_beta_0_to_fp16 = const()[name = string("input_589_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288464192)))]; - fp16 input_589_epsilon_0_to_fp16 = const()[name = string("input_589_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_589_cast_fp16 = batch_norm(beta = input_589_beta_0_to_fp16, epsilon = input_589_epsilon_0_to_fp16, gamma = input_589_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_221_cast_fp16)[name = string("input_589_cast_fp16")]; - string var_8570_pad_type_0 = const()[name = string("op_8570_pad_type_0"), val = string("valid")]; - tensor var_8570_strides_0 = const()[name = string("op_8570_strides_0"), val = tensor([1, 1])]; - tensor var_8570_pad_0 = const()[name = string("op_8570_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8570_dilations_0 = const()[name = string("op_8570_dilations_0"), val = tensor([1, 1])]; - int32 var_8570_groups_0 = const()[name = string("op_8570_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288466304))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290039232))))[name = string("layers_22_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8570_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8570_dilations_0, groups = var_8570_groups_0, pad = var_8570_pad_0, pad_type = var_8570_pad_type_0, strides = var_8570_strides_0, weight = layers_22_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_589_cast_fp16)[name = string("op_8570_cast_fp16")]; - string var_8576_pad_type_0 = const()[name = string("op_8576_pad_type_0"), val = string("valid")]; - tensor var_8576_strides_0 = const()[name = string("op_8576_strides_0"), val = tensor([1, 1])]; - tensor var_8576_pad_0 = const()[name = string("op_8576_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8576_dilations_0 = const()[name = string("op_8576_dilations_0"), val = tensor([1, 1])]; - int32 var_8576_groups_0 = const()[name = string("op_8576_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290080960))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290043392))))[name = string("layers_22_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8576_cast_fp16 = conv(dilations = var_8576_dilations_0, groups = var_8576_groups_0, pad = var_8576_pad_0, pad_type = var_8576_pad_type_0, strides = var_8576_strides_0, weight = layers_22_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_589_cast_fp16)[name = string("op_8576_cast_fp16")]; - tensor input_591_cast_fp16 = add(x = var_8570_cast_fp16, y = var_8576_cast_fp16)[name = string("input_591_cast_fp16")]; - tensor input_593_cast_fp16 = silu(x = input_591_cast_fp16)[name = string("input_593_cast_fp16")]; - string var_8587_pad_type_0 = const()[name = string("op_8587_pad_type_0"), val = string("valid")]; - tensor var_8587_strides_0 = const()[name = string("op_8587_strides_0"), val = tensor([1, 1])]; - tensor var_8587_pad_0 = const()[name = string("op_8587_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8587_dilations_0 = const()[name = string("op_8587_dilations_0"), val = tensor([1, 1])]; - int32 var_8587_groups_0 = const()[name = string("op_8587_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290605312))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292178240))))[name = string("layers_22_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8587_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8587_dilations_0, groups = var_8587_groups_0, pad = var_8587_pad_0, pad_type = var_8587_pad_type_0, strides = var_8587_strides_0, weight = layers_22_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_593_cast_fp16)[name = string("op_8587_cast_fp16")]; - string var_8593_pad_type_0 = const()[name = string("op_8593_pad_type_0"), val = string("valid")]; - tensor var_8593_strides_0 = const()[name = string("op_8593_strides_0"), val = tensor([1, 1])]; - tensor var_8593_pad_0 = const()[name = string("op_8593_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8593_dilations_0 = const()[name = string("op_8593_dilations_0"), val = tensor([1, 1])]; - int32 var_8593_groups_0 = const()[name = string("op_8593_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292228480))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292179328))))[name = string("layers_22_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8593_cast_fp16 = conv(dilations = var_8593_dilations_0, groups = var_8593_groups_0, pad = var_8593_pad_0, pad_type = var_8593_pad_type_0, strides = var_8593_strides_0, weight = layers_22_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_593_cast_fp16)[name = string("op_8593_cast_fp16")]; - tensor x_135_cast_fp16 = add(x = var_8587_cast_fp16, y = var_8593_cast_fp16)[name = string("x_135_cast_fp16")]; - fp16 var_8595_to_fp16 = const()[name = string("op_8595_to_fp16"), val = fp16(0x1p-1)]; - tensor var_8596_cast_fp16 = mul(x = x_135_cast_fp16, y = var_8595_to_fp16)[name = string("op_8596_cast_fp16")]; - tensor inputs_223_cast_fp16 = add(x = inputs_221_cast_fp16, y = var_8596_cast_fp16)[name = string("inputs_223_cast_fp16")]; - tensor out_223_axes_0 = const()[name = string("out_223_axes_0"), val = tensor([1])]; - fp16 var_8606_to_fp16 = const()[name = string("op_8606_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_223_cast_fp16 = layer_norm(axes = out_223_axes_0, epsilon = var_8606_to_fp16, x = inputs_223_cast_fp16)[name = string("out_223_cast_fp16")]; - tensor obj_91_gamma_0_to_fp16 = const()[name = string("obj_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292752832)))]; - tensor obj_91_beta_0_to_fp16 = const()[name = string("obj_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292754944)))]; - fp16 obj_91_epsilon_0_to_fp16 = const()[name = string("obj_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_91_cast_fp16 = batch_norm(beta = obj_91_beta_0_to_fp16, epsilon = obj_91_epsilon_0_to_fp16, gamma = obj_91_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_223_cast_fp16)[name = string("obj_91_cast_fp16")]; - string var_8631_pad_type_0 = const()[name = string("op_8631_pad_type_0"), val = string("valid")]; - tensor var_8631_strides_0 = const()[name = string("op_8631_strides_0"), val = tensor([1, 1])]; - tensor var_8631_pad_0 = const()[name = string("op_8631_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8631_dilations_0 = const()[name = string("op_8631_dilations_0"), val = tensor([1, 1])]; - int32 var_8631_groups_0 = const()[name = string("op_8631_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292757056))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293150336))))[name = string("layers_22_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8631_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8631_dilations_0, groups = var_8631_groups_0, pad = var_8631_pad_0, pad_type = var_8631_pad_type_0, strides = var_8631_strides_0, weight = layers_22_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_91_cast_fp16)[name = string("op_8631_cast_fp16")]; - string var_8637_pad_type_0 = const()[name = string("op_8637_pad_type_0"), val = string("valid")]; - tensor var_8637_strides_0 = const()[name = string("op_8637_strides_0"), val = tensor([1, 1])]; - tensor var_8637_pad_0 = const()[name = string("op_8637_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8637_dilations_0 = const()[name = string("op_8637_dilations_0"), val = tensor([1, 1])]; - int32 var_8637_groups_0 = const()[name = string("op_8637_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293159296))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293151424))))[name = string("layers_22_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8637_cast_fp16 = conv(dilations = var_8637_dilations_0, groups = var_8637_groups_0, pad = var_8637_pad_0, pad_type = var_8637_pad_type_0, strides = var_8637_strides_0, weight = layers_22_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_91_cast_fp16)[name = string("op_8637_cast_fp16")]; - tensor query_89_cast_fp16 = add(x = var_8631_cast_fp16, y = var_8637_cast_fp16)[name = string("query_89_cast_fp16")]; - string var_8646_pad_type_0 = const()[name = string("op_8646_pad_type_0"), val = string("valid")]; - tensor var_8646_strides_0 = const()[name = string("op_8646_strides_0"), val = tensor([1, 1])]; - tensor var_8646_pad_0 = const()[name = string("op_8646_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8646_dilations_0 = const()[name = string("op_8646_dilations_0"), val = tensor([1, 1])]; - int32 var_8646_groups_0 = const()[name = string("op_8646_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293290432))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293683712))))[name = string("layers_22_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8646_cast_fp16 = conv(dilations = var_8646_dilations_0, groups = var_8646_groups_0, pad = var_8646_pad_0, pad_type = var_8646_pad_type_0, strides = var_8646_strides_0, weight = layers_22_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_91_cast_fp16)[name = string("op_8646_cast_fp16")]; - string var_8652_pad_type_0 = const()[name = string("op_8652_pad_type_0"), val = string("valid")]; - tensor var_8652_strides_0 = const()[name = string("op_8652_strides_0"), val = tensor([1, 1])]; - tensor var_8652_pad_0 = const()[name = string("op_8652_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8652_dilations_0 = const()[name = string("op_8652_dilations_0"), val = tensor([1, 1])]; - int32 var_8652_groups_0 = const()[name = string("op_8652_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293694912))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293684800))))[name = string("layers_22_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8652_cast_fp16 = conv(dilations = var_8652_dilations_0, groups = var_8652_groups_0, pad = var_8652_pad_0, pad_type = var_8652_pad_type_0, strides = var_8652_strides_0, weight = layers_22_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_91_cast_fp16)[name = string("op_8652_cast_fp16")]; - tensor key_45_cast_fp16 = add(x = var_8646_cast_fp16, y = var_8652_cast_fp16)[name = string("key_45_cast_fp16")]; - string var_8662_pad_type_0 = const()[name = string("op_8662_pad_type_0"), val = string("valid")]; - tensor var_8662_strides_0 = const()[name = string("op_8662_strides_0"), val = tensor([1, 1])]; - tensor var_8662_pad_0 = const()[name = string("op_8662_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8662_dilations_0 = const()[name = string("op_8662_dilations_0"), val = tensor([1, 1])]; - int32 var_8662_groups_0 = const()[name = string("op_8662_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293826048))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294219328))))[name = string("layers_22_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8662_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8662_dilations_0, groups = var_8662_groups_0, pad = var_8662_pad_0, pad_type = var_8662_pad_type_0, strides = var_8662_strides_0, weight = layers_22_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_91_cast_fp16)[name = string("op_8662_cast_fp16")]; - string var_8668_pad_type_0 = const()[name = string("op_8668_pad_type_0"), val = string("valid")]; - tensor var_8668_strides_0 = const()[name = string("op_8668_strides_0"), val = tensor([1, 1])]; - tensor var_8668_pad_0 = const()[name = string("op_8668_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8668_dilations_0 = const()[name = string("op_8668_dilations_0"), val = tensor([1, 1])]; - int32 var_8668_groups_0 = const()[name = string("op_8668_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294228032))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294220416))))[name = string("layers_22_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8668_cast_fp16 = conv(dilations = var_8668_dilations_0, groups = var_8668_groups_0, pad = var_8668_pad_0, pad_type = var_8668_pad_type_0, strides = var_8668_strides_0, weight = layers_22_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_91_cast_fp16)[name = string("op_8668_cast_fp16")]; - tensor value_45_cast_fp16 = add(x = var_8662_cast_fp16, y = var_8668_cast_fp16)[name = string("value_45_cast_fp16")]; - tensor var_8671_to_fp16 = const()[name = string("op_8671_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294359168)))]; - tensor query_91_cast_fp16 = add(x = query_89_cast_fp16, y = var_8671_to_fp16)[name = string("query_91_cast_fp16")]; - tensor var_8674_to_fp16 = const()[name = string("op_8674_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294361280)))]; - tensor q_with_bias_v_45_cast_fp16 = add(x = query_89_cast_fp16, y = var_8674_to_fp16)[name = string("q_with_bias_v_45_cast_fp16")]; - string var_8684_pad_type_0 = const()[name = string("op_8684_pad_type_0"), val = string("valid")]; - tensor var_8684_strides_0 = const()[name = string("op_8684_strides_0"), val = tensor([1, 1])]; - tensor var_8684_pad_0 = const()[name = string("op_8684_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8684_dilations_0 = const()[name = string("op_8684_dilations_0"), val = tensor([1, 1])]; - int32 var_8684_groups_0 = const()[name = string("op_8684_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294363392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294756672))))[name = string("layers_22_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_8684_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8684_dilations_0, groups = var_8684_groups_0, pad = var_8684_pad_0, pad_type = var_8684_pad_type_0, strides = var_8684_strides_0, weight = layers_22_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_8684_cast_fp16")]; - string var_8690_pad_type_0 = const()[name = string("op_8690_pad_type_0"), val = string("valid")]; - tensor var_8690_strides_0 = const()[name = string("op_8690_strides_0"), val = tensor([1, 1])]; - tensor var_8690_pad_0 = const()[name = string("op_8690_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8690_dilations_0 = const()[name = string("op_8690_dilations_0"), val = tensor([1, 1])]; - int32 var_8690_groups_0 = const()[name = string("op_8690_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294781888))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294757760))))[name = string("layers_22_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8690_cast_fp16 = conv(dilations = var_8690_dilations_0, groups = var_8690_groups_0, pad = var_8690_pad_0, pad_type = var_8690_pad_type_0, strides = var_8690_strides_0, weight = layers_22_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_8690_cast_fp16")]; - tensor p_45_cast_fp16 = add(x = var_8684_cast_fp16, y = var_8690_cast_fp16)[name = string("p_45_cast_fp16")]; - tensor var_8694 = const()[name = string("op_8694"), val = tensor([1, 8, 128, 188])]; - tensor var_8695_cast_fp16 = reshape(shape = var_8694, x = q_with_bias_v_45_cast_fp16)[name = string("op_8695_cast_fp16")]; - tensor var_8696 = const()[name = string("op_8696"), val = tensor([1, 8, 128, -1])]; - tensor var_8697_cast_fp16 = reshape(shape = var_8696, x = p_45_cast_fp16)[name = string("op_8697_cast_fp16")]; - bool matrix_bd_177_transpose_x_0 = const()[name = string("matrix_bd_177_transpose_x_0"), val = bool(true)]; - bool matrix_bd_177_transpose_y_0 = const()[name = string("matrix_bd_177_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_177_cast_fp16 = matmul(transpose_x = matrix_bd_177_transpose_x_0, transpose_y = matrix_bd_177_transpose_y_0, x = var_8695_cast_fp16, y = var_8697_cast_fp16)[name = string("matrix_bd_177_cast_fp16")]; - tensor matrix_bd_179_pad_0 = const()[name = string("matrix_bd_179_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_179_mode_0 = const()[name = string("matrix_bd_179_mode_0"), val = string("constant")]; - fp16 const_252_to_fp16 = const()[name = string("const_252_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_179_cast_fp16 = pad(constant_val = const_252_to_fp16, mode = matrix_bd_179_mode_0, pad = matrix_bd_179_pad_0, x = matrix_bd_177_cast_fp16)[name = string("matrix_bd_179_cast_fp16")]; - tensor var_8706 = const()[name = string("op_8706"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_181_cast_fp16 = reshape(shape = var_8706, x = matrix_bd_179_cast_fp16)[name = string("matrix_bd_181_cast_fp16")]; - tensor var_8710_begin_0 = const()[name = string("op_8710_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_8710_end_0 = const()[name = string("op_8710_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_8710_end_mask_0 = const()[name = string("op_8710_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_8710_cast_fp16 = slice_by_index(begin = var_8710_begin_0, end = var_8710_end_0, end_mask = var_8710_end_mask_0, x = matrix_bd_181_cast_fp16)[name = string("op_8710_cast_fp16")]; - tensor var_8711 = const()[name = string("op_8711"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_183_cast_fp16 = reshape(shape = var_8711, x = var_8710_cast_fp16)[name = string("matrix_bd_183_cast_fp16")]; - tensor var_8716_begin_0 = const()[name = string("op_8716_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8716_end_0 = const()[name = string("op_8716_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_8716_end_mask_0 = const()[name = string("op_8716_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_8716_cast_fp16 = slice_by_index(begin = var_8716_begin_0, end = var_8716_end_0, end_mask = var_8716_end_mask_0, x = matrix_bd_183_cast_fp16)[name = string("op_8716_cast_fp16")]; - fp16 var_8717_to_fp16 = const()[name = string("op_8717_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_45_cast_fp16 = mul(x = var_8716_cast_fp16, y = var_8717_to_fp16)[name = string("qk_mask_45_cast_fp16")]; - tensor var_8721 = const()[name = string("op_8721"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_45_cast_fp16 = reshape(shape = var_8721, x = query_91_cast_fp16)[name = string("mh_q_45_cast_fp16")]; - fp16 var_8723_to_fp16 = const()[name = string("op_8723_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_8724_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_8723_to_fp16)[name = string("op_8724_cast_fp16")]; - tensor var_8727 = const()[name = string("op_8727"), val = tensor([1, 8, 128, 188])]; - tensor var_8728_cast_fp16 = reshape(shape = var_8727, x = key_45_cast_fp16)[name = string("op_8728_cast_fp16")]; - bool mh_w_89_transpose_x_0 = const()[name = string("mh_w_89_transpose_x_0"), val = bool(true)]; - bool mh_w_89_transpose_y_0 = const()[name = string("mh_w_89_transpose_y_0"), val = bool(false)]; - tensor mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_8724_cast_fp16, y = var_8728_cast_fp16)[name = string("mh_w_89_cast_fp16")]; - tensor mh_w_91_cast_fp16 = add(x = mh_w_89_cast_fp16, y = qk_mask_45_cast_fp16)[name = string("mh_w_91_cast_fp16")]; - tensor var_8732_cast_fp16 = softmax(axis = var_8519, x = mh_w_91_cast_fp16)[name = string("op_8732_cast_fp16")]; - tensor var_8733 = const()[name = string("op_8733"), val = tensor([1, 8, 128, 188])]; - tensor var_8734_cast_fp16 = reshape(shape = var_8733, x = value_45_cast_fp16)[name = string("op_8734_cast_fp16")]; - bool attn_45_transpose_x_0 = const()[name = string("attn_45_transpose_x_0"), val = bool(false)]; - bool attn_45_transpose_y_0 = const()[name = string("attn_45_transpose_y_0"), val = bool(true)]; - tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_8734_cast_fp16, y = var_8732_cast_fp16)[name = string("attn_45_cast_fp16")]; - tensor var_8737 = const()[name = string("op_8737"), val = tensor([1, 1024, 1, 188])]; - tensor input_595_cast_fp16 = reshape(shape = var_8737, x = attn_45_cast_fp16)[name = string("input_595_cast_fp16")]; - string var_8747_pad_type_0 = const()[name = string("op_8747_pad_type_0"), val = string("valid")]; - tensor var_8747_strides_0 = const()[name = string("op_8747_strides_0"), val = tensor([1, 1])]; - tensor var_8747_pad_0 = const()[name = string("op_8747_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8747_dilations_0 = const()[name = string("op_8747_dilations_0"), val = tensor([1, 1])]; - int32 var_8747_groups_0 = const()[name = string("op_8747_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294913024))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295306304))))[name = string("layers_22_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_8747_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8747_dilations_0, groups = var_8747_groups_0, pad = var_8747_pad_0, pad_type = var_8747_pad_type_0, strides = var_8747_strides_0, weight = layers_22_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_595_cast_fp16)[name = string("op_8747_cast_fp16")]; - string var_8753_pad_type_0 = const()[name = string("op_8753_pad_type_0"), val = string("valid")]; - tensor var_8753_strides_0 = const()[name = string("op_8753_strides_0"), val = tensor([1, 1])]; - tensor var_8753_pad_0 = const()[name = string("op_8753_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8753_dilations_0 = const()[name = string("op_8753_dilations_0"), val = tensor([1, 1])]; - int32 var_8753_groups_0 = const()[name = string("op_8753_groups_0"), val = int32(1)]; - tensor layers_22_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295317632))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295307392))))[name = string("layers_22_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8753_cast_fp16 = conv(dilations = var_8753_dilations_0, groups = var_8753_groups_0, pad = var_8753_pad_0, pad_type = var_8753_pad_type_0, strides = var_8753_strides_0, weight = layers_22_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_595_cast_fp16)[name = string("op_8753_cast_fp16")]; - tensor obj_93_cast_fp16 = add(x = var_8747_cast_fp16, y = var_8753_cast_fp16)[name = string("obj_93_cast_fp16")]; - tensor inputs_225_cast_fp16 = add(x = inputs_223_cast_fp16, y = obj_93_cast_fp16)[name = string("inputs_225_cast_fp16")]; - tensor out_225_axes_0 = const()[name = string("out_225_axes_0"), val = tensor([1])]; - fp16 var_8764_to_fp16 = const()[name = string("op_8764_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_225_cast_fp16 = layer_norm(axes = out_225_axes_0, epsilon = var_8764_to_fp16, x = inputs_225_cast_fp16)[name = string("out_225_cast_fp16")]; - tensor input_597_gamma_0_to_fp16 = const()[name = string("input_597_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295448768)))]; - tensor input_597_beta_0_to_fp16 = const()[name = string("input_597_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295450880)))]; - fp16 input_597_epsilon_0_to_fp16 = const()[name = string("input_597_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_597_cast_fp16 = batch_norm(beta = input_597_beta_0_to_fp16, epsilon = input_597_epsilon_0_to_fp16, gamma = input_597_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_225_cast_fp16)[name = string("input_597_cast_fp16")]; - string var_8785_pad_type_0 = const()[name = string("op_8785_pad_type_0"), val = string("valid")]; - tensor var_8785_strides_0 = const()[name = string("op_8785_strides_0"), val = tensor([1, 1])]; - tensor var_8785_pad_0 = const()[name = string("op_8785_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8785_dilations_0 = const()[name = string("op_8785_dilations_0"), val = tensor([1, 1])]; - int32 var_8785_groups_0 = const()[name = string("op_8785_groups_0"), val = int32(1)]; - tensor layers_22_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295452992))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296239488))))[name = string("layers_22_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8785_cast_fp16 = conv(dilations = var_8785_dilations_0, groups = var_8785_groups_0, pad = var_8785_pad_0, pad_type = var_8785_pad_type_0, strides = var_8785_strides_0, weight = layers_22_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_597_cast_fp16)[name = string("op_8785_cast_fp16")]; - string var_8791_pad_type_0 = const()[name = string("op_8791_pad_type_0"), val = string("valid")]; - tensor var_8791_strides_0 = const()[name = string("op_8791_strides_0"), val = tensor([1, 1])]; - tensor var_8791_pad_0 = const()[name = string("op_8791_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8791_dilations_0 = const()[name = string("op_8791_dilations_0"), val = tensor([1, 1])]; - int32 var_8791_groups_0 = const()[name = string("op_8791_groups_0"), val = int32(1)]; - tensor layers_22_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296261568))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296241600))))[name = string("layers_22_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8791_cast_fp16 = conv(dilations = var_8791_dilations_0, groups = var_8791_groups_0, pad = var_8791_pad_0, pad_type = var_8791_pad_type_0, strides = var_8791_strides_0, weight = layers_22_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_597_cast_fp16)[name = string("op_8791_cast_fp16")]; - tensor input_599_cast_fp16 = add(x = var_8785_cast_fp16, y = var_8791_cast_fp16)[name = string("input_599_cast_fp16")]; - int32 input_601_split_num_splits_0 = const()[name = string("input_601_split_num_splits_0"), val = int32(2)]; - int32 input_601_split_axis_0 = const()[name = string("input_601_split_axis_0"), val = int32(1)]; - tensor input_601_split_cast_fp16_0, tensor input_601_split_cast_fp16_1 = split(axis = input_601_split_axis_0, num_splits = input_601_split_num_splits_0, x = input_599_cast_fp16)[name = string("input_601_split_cast_fp16")]; - tensor input_601_split_1_sigmoid_cast_fp16 = sigmoid(x = input_601_split_cast_fp16_1)[name = string("input_601_split_1_sigmoid_cast_fp16")]; - tensor input_601_cast_fp16 = mul(x = input_601_split_cast_fp16_0, y = input_601_split_1_sigmoid_cast_fp16)[name = string("input_601_cast_fp16")]; - string input_603_pad_type_0 = const()[name = string("input_603_pad_type_0"), val = string("custom")]; - tensor input_603_pad_0 = const()[name = string("input_603_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_603_groups_0 = const()[name = string("input_603_groups_0"), val = int32(1024)]; - tensor input_603_strides_0 = const()[name = string("input_603_strides_0"), val = tensor([1, 1])]; - tensor input_603_dilations_0 = const()[name = string("input_603_dilations_0"), val = tensor([1, 1])]; - tensor const_312_to_fp16 = const()[name = string("const_312_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296523776)))]; - tensor const_313_to_fp16 = const()[name = string("const_313_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296542272)))]; - tensor input_605_cast_fp16 = conv(bias = const_313_to_fp16, dilations = input_603_dilations_0, groups = input_603_groups_0, pad = input_603_pad_0, pad_type = input_603_pad_type_0, strides = input_603_strides_0, weight = const_312_to_fp16, x = input_601_cast_fp16)[name = string("input_605_cast_fp16")]; - tensor input_607_cast_fp16 = silu(x = input_605_cast_fp16)[name = string("input_607_cast_fp16")]; - string var_8813_pad_type_0 = const()[name = string("op_8813_pad_type_0"), val = string("valid")]; - tensor var_8813_strides_0 = const()[name = string("op_8813_strides_0"), val = tensor([1, 1])]; - tensor var_8813_pad_0 = const()[name = string("op_8813_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8813_dilations_0 = const()[name = string("op_8813_dilations_0"), val = tensor([1, 1])]; - int32 var_8813_groups_0 = const()[name = string("op_8813_groups_0"), val = int32(1)]; - tensor layers_22_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296544384))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296937664))))[name = string("layers_22_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8813_cast_fp16 = conv(dilations = var_8813_dilations_0, groups = var_8813_groups_0, pad = var_8813_pad_0, pad_type = var_8813_pad_type_0, strides = var_8813_strides_0, weight = layers_22_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_607_cast_fp16)[name = string("op_8813_cast_fp16")]; - string var_8819_pad_type_0 = const()[name = string("op_8819_pad_type_0"), val = string("valid")]; - tensor var_8819_strides_0 = const()[name = string("op_8819_strides_0"), val = tensor([1, 1])]; - tensor var_8819_pad_0 = const()[name = string("op_8819_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8819_dilations_0 = const()[name = string("op_8819_dilations_0"), val = tensor([1, 1])]; - int32 var_8819_groups_0 = const()[name = string("op_8819_groups_0"), val = int32(1)]; - tensor layers_22_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296950336))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296938752))))[name = string("layers_22_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8819_cast_fp16 = conv(dilations = var_8819_dilations_0, groups = var_8819_groups_0, pad = var_8819_pad_0, pad_type = var_8819_pad_type_0, strides = var_8819_strides_0, weight = layers_22_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_607_cast_fp16)[name = string("op_8819_cast_fp16")]; - tensor x_137_cast_fp16 = add(x = var_8813_cast_fp16, y = var_8819_cast_fp16)[name = string("x_137_cast_fp16")]; - tensor inputs_227_cast_fp16 = add(x = inputs_225_cast_fp16, y = x_137_cast_fp16)[name = string("inputs_227_cast_fp16")]; - tensor out_227_axes_0 = const()[name = string("out_227_axes_0"), val = tensor([1])]; - fp16 var_8830_to_fp16 = const()[name = string("op_8830_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_227_cast_fp16 = layer_norm(axes = out_227_axes_0, epsilon = var_8830_to_fp16, x = inputs_227_cast_fp16)[name = string("out_227_cast_fp16")]; - tensor input_609_gamma_0_to_fp16 = const()[name = string("input_609_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297081472)))]; - tensor input_609_beta_0_to_fp16 = const()[name = string("input_609_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297083584)))]; - fp16 input_609_epsilon_0_to_fp16 = const()[name = string("input_609_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_609_cast_fp16 = batch_norm(beta = input_609_beta_0_to_fp16, epsilon = input_609_epsilon_0_to_fp16, gamma = input_609_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_227_cast_fp16)[name = string("input_609_cast_fp16")]; - string var_8850_pad_type_0 = const()[name = string("op_8850_pad_type_0"), val = string("valid")]; - tensor var_8850_strides_0 = const()[name = string("op_8850_strides_0"), val = tensor([1, 1])]; - tensor var_8850_pad_0 = const()[name = string("op_8850_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8850_dilations_0 = const()[name = string("op_8850_dilations_0"), val = tensor([1, 1])]; - int32 var_8850_groups_0 = const()[name = string("op_8850_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297085696))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298658624))))[name = string("layers_22_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8850_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8850_dilations_0, groups = var_8850_groups_0, pad = var_8850_pad_0, pad_type = var_8850_pad_type_0, strides = var_8850_strides_0, weight = layers_22_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_609_cast_fp16)[name = string("op_8850_cast_fp16")]; - string var_8856_pad_type_0 = const()[name = string("op_8856_pad_type_0"), val = string("valid")]; - tensor var_8856_strides_0 = const()[name = string("op_8856_strides_0"), val = tensor([1, 1])]; - tensor var_8856_pad_0 = const()[name = string("op_8856_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8856_dilations_0 = const()[name = string("op_8856_dilations_0"), val = tensor([1, 1])]; - int32 var_8856_groups_0 = const()[name = string("op_8856_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298699648))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298662784))))[name = string("layers_22_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8856_cast_fp16 = conv(dilations = var_8856_dilations_0, groups = var_8856_groups_0, pad = var_8856_pad_0, pad_type = var_8856_pad_type_0, strides = var_8856_strides_0, weight = layers_22_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_609_cast_fp16)[name = string("op_8856_cast_fp16")]; - tensor input_611_cast_fp16 = add(x = var_8850_cast_fp16, y = var_8856_cast_fp16)[name = string("input_611_cast_fp16")]; - tensor input_613_cast_fp16 = silu(x = input_611_cast_fp16)[name = string("input_613_cast_fp16")]; - string var_8867_pad_type_0 = const()[name = string("op_8867_pad_type_0"), val = string("valid")]; - tensor var_8867_strides_0 = const()[name = string("op_8867_strides_0"), val = tensor([1, 1])]; - tensor var_8867_pad_0 = const()[name = string("op_8867_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8867_dilations_0 = const()[name = string("op_8867_dilations_0"), val = tensor([1, 1])]; - int32 var_8867_groups_0 = const()[name = string("op_8867_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299224000))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300796928))))[name = string("layers_22_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8867_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8867_dilations_0, groups = var_8867_groups_0, pad = var_8867_pad_0, pad_type = var_8867_pad_type_0, strides = var_8867_strides_0, weight = layers_22_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_613_cast_fp16)[name = string("op_8867_cast_fp16")]; - string var_8873_pad_type_0 = const()[name = string("op_8873_pad_type_0"), val = string("valid")]; - tensor var_8873_strides_0 = const()[name = string("op_8873_strides_0"), val = tensor([1, 1])]; - tensor var_8873_pad_0 = const()[name = string("op_8873_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8873_dilations_0 = const()[name = string("op_8873_dilations_0"), val = tensor([1, 1])]; - int32 var_8873_groups_0 = const()[name = string("op_8873_groups_0"), val = int32(1)]; - tensor layers_22_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300881280))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300798016))))[name = string("layers_22_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8873_cast_fp16 = conv(dilations = var_8873_dilations_0, groups = var_8873_groups_0, pad = var_8873_pad_0, pad_type = var_8873_pad_type_0, strides = var_8873_strides_0, weight = layers_22_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_613_cast_fp16)[name = string("op_8873_cast_fp16")]; - tensor x_139_cast_fp16 = add(x = var_8867_cast_fp16, y = var_8873_cast_fp16)[name = string("x_139_cast_fp16")]; - fp16 var_8875_to_fp16 = const()[name = string("op_8875_to_fp16"), val = fp16(0x1p-1)]; - tensor var_8876_cast_fp16 = mul(x = x_139_cast_fp16, y = var_8875_to_fp16)[name = string("op_8876_cast_fp16")]; - tensor inputs_229_cast_fp16 = add(x = inputs_227_cast_fp16, y = var_8876_cast_fp16)[name = string("inputs_229_cast_fp16")]; - tensor out_229_axes_0 = const()[name = string("out_229_axes_0"), val = tensor([1])]; - fp16 var_8886_to_fp16 = const()[name = string("op_8886_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_229_cast_fp16 = layer_norm(axes = out_229_axes_0, epsilon = var_8886_to_fp16, x = inputs_229_cast_fp16)[name = string("out_229_cast_fp16")]; - tensor inputs_231_gamma_0_to_fp16 = const()[name = string("inputs_231_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301405632)))]; - tensor inputs_231_beta_0_to_fp16 = const()[name = string("inputs_231_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301407744)))]; - fp16 inputs_231_epsilon_0_to_fp16 = const()[name = string("inputs_231_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor inputs_231_cast_fp16 = batch_norm(beta = inputs_231_beta_0_to_fp16, epsilon = inputs_231_epsilon_0_to_fp16, gamma = inputs_231_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_229_cast_fp16)[name = string("inputs_231_cast_fp16")]; - int32 var_8900 = const()[name = string("op_8900"), val = int32(3)]; - tensor out_231_axes_0 = const()[name = string("out_231_axes_0"), val = tensor([1])]; - fp16 var_8931_to_fp16 = const()[name = string("op_8931_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_231_cast_fp16 = layer_norm(axes = out_231_axes_0, epsilon = var_8931_to_fp16, x = inputs_231_cast_fp16)[name = string("out_231_cast_fp16")]; - tensor input_615_gamma_0_to_fp16 = const()[name = string("input_615_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301409856)))]; - tensor input_615_beta_0_to_fp16 = const()[name = string("input_615_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301411968)))]; - fp16 input_615_epsilon_0_to_fp16 = const()[name = string("input_615_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_615_cast_fp16 = batch_norm(beta = input_615_beta_0_to_fp16, epsilon = input_615_epsilon_0_to_fp16, gamma = input_615_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_231_cast_fp16)[name = string("input_615_cast_fp16")]; - string var_8951_pad_type_0 = const()[name = string("op_8951_pad_type_0"), val = string("valid")]; - tensor var_8951_strides_0 = const()[name = string("op_8951_strides_0"), val = tensor([1, 1])]; - tensor var_8951_pad_0 = const()[name = string("op_8951_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8951_dilations_0 = const()[name = string("op_8951_dilations_0"), val = tensor([1, 1])]; - int32 var_8951_groups_0 = const()[name = string("op_8951_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301414080))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302987008))))[name = string("layers_23_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_8951_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8951_dilations_0, groups = var_8951_groups_0, pad = var_8951_pad_0, pad_type = var_8951_pad_type_0, strides = var_8951_strides_0, weight = layers_23_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_615_cast_fp16)[name = string("op_8951_cast_fp16")]; - string var_8957_pad_type_0 = const()[name = string("op_8957_pad_type_0"), val = string("valid")]; - tensor var_8957_strides_0 = const()[name = string("op_8957_strides_0"), val = tensor([1, 1])]; - tensor var_8957_pad_0 = const()[name = string("op_8957_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8957_dilations_0 = const()[name = string("op_8957_dilations_0"), val = tensor([1, 1])]; - int32 var_8957_groups_0 = const()[name = string("op_8957_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303027328))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302991168))))[name = string("layers_23_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8957_cast_fp16 = conv(dilations = var_8957_dilations_0, groups = var_8957_groups_0, pad = var_8957_pad_0, pad_type = var_8957_pad_type_0, strides = var_8957_strides_0, weight = layers_23_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_615_cast_fp16)[name = string("op_8957_cast_fp16")]; - tensor input_617_cast_fp16 = add(x = var_8951_cast_fp16, y = var_8957_cast_fp16)[name = string("input_617_cast_fp16")]; - tensor input_619_cast_fp16 = silu(x = input_617_cast_fp16)[name = string("input_619_cast_fp16")]; - string var_8968_pad_type_0 = const()[name = string("op_8968_pad_type_0"), val = string("valid")]; - tensor var_8968_strides_0 = const()[name = string("op_8968_strides_0"), val = tensor([1, 1])]; - tensor var_8968_pad_0 = const()[name = string("op_8968_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8968_dilations_0 = const()[name = string("op_8968_dilations_0"), val = tensor([1, 1])]; - int32 var_8968_groups_0 = const()[name = string("op_8968_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303551680))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305124608))))[name = string("layers_23_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_8968_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8968_dilations_0, groups = var_8968_groups_0, pad = var_8968_pad_0, pad_type = var_8968_pad_type_0, strides = var_8968_strides_0, weight = layers_23_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_619_cast_fp16)[name = string("op_8968_cast_fp16")]; - string var_8974_pad_type_0 = const()[name = string("op_8974_pad_type_0"), val = string("valid")]; - tensor var_8974_strides_0 = const()[name = string("op_8974_strides_0"), val = tensor([1, 1])]; - tensor var_8974_pad_0 = const()[name = string("op_8974_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_8974_dilations_0 = const()[name = string("op_8974_dilations_0"), val = tensor([1, 1])]; - int32 var_8974_groups_0 = const()[name = string("op_8974_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305211904))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305125696))))[name = string("layers_23_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_8974_cast_fp16 = conv(dilations = var_8974_dilations_0, groups = var_8974_groups_0, pad = var_8974_pad_0, pad_type = var_8974_pad_type_0, strides = var_8974_strides_0, weight = layers_23_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_619_cast_fp16)[name = string("op_8974_cast_fp16")]; - tensor x_141_cast_fp16 = add(x = var_8968_cast_fp16, y = var_8974_cast_fp16)[name = string("x_141_cast_fp16")]; - fp16 var_8976_to_fp16 = const()[name = string("op_8976_to_fp16"), val = fp16(0x1p-1)]; - tensor var_8977_cast_fp16 = mul(x = x_141_cast_fp16, y = var_8976_to_fp16)[name = string("op_8977_cast_fp16")]; - tensor inputs_233_cast_fp16 = add(x = inputs_231_cast_fp16, y = var_8977_cast_fp16)[name = string("inputs_233_cast_fp16")]; - tensor out_233_axes_0 = const()[name = string("out_233_axes_0"), val = tensor([1])]; - fp16 var_8987_to_fp16 = const()[name = string("op_8987_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_233_cast_fp16 = layer_norm(axes = out_233_axes_0, epsilon = var_8987_to_fp16, x = inputs_233_cast_fp16)[name = string("out_233_cast_fp16")]; - tensor obj_95_gamma_0_to_fp16 = const()[name = string("obj_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305736256)))]; - tensor obj_95_beta_0_to_fp16 = const()[name = string("obj_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305738368)))]; - fp16 obj_95_epsilon_0_to_fp16 = const()[name = string("obj_95_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor obj_95_cast_fp16 = batch_norm(beta = obj_95_beta_0_to_fp16, epsilon = obj_95_epsilon_0_to_fp16, gamma = obj_95_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_233_cast_fp16)[name = string("obj_95_cast_fp16")]; - string var_9012_pad_type_0 = const()[name = string("op_9012_pad_type_0"), val = string("valid")]; - tensor var_9012_strides_0 = const()[name = string("op_9012_strides_0"), val = tensor([1, 1])]; - tensor var_9012_pad_0 = const()[name = string("op_9012_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9012_dilations_0 = const()[name = string("op_9012_dilations_0"), val = tensor([1, 1])]; - int32 var_9012_groups_0 = const()[name = string("op_9012_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305740480))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306133760))))[name = string("layers_23_self_attn_q_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_9012_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9012_dilations_0, groups = var_9012_groups_0, pad = var_9012_pad_0, pad_type = var_9012_pad_type_0, strides = var_9012_strides_0, weight = layers_23_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_95_cast_fp16)[name = string("op_9012_cast_fp16")]; - string var_9018_pad_type_0 = const()[name = string("op_9018_pad_type_0"), val = string("valid")]; - tensor var_9018_strides_0 = const()[name = string("op_9018_strides_0"), val = tensor([1, 1])]; - tensor var_9018_pad_0 = const()[name = string("op_9018_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9018_dilations_0 = const()[name = string("op_9018_dilations_0"), val = tensor([1, 1])]; - int32 var_9018_groups_0 = const()[name = string("op_9018_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306142528))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306134848))))[name = string("layers_23_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9018_cast_fp16 = conv(dilations = var_9018_dilations_0, groups = var_9018_groups_0, pad = var_9018_pad_0, pad_type = var_9018_pad_type_0, strides = var_9018_strides_0, weight = layers_23_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_95_cast_fp16)[name = string("op_9018_cast_fp16")]; - tensor query_93_cast_fp16 = add(x = var_9012_cast_fp16, y = var_9018_cast_fp16)[name = string("query_93_cast_fp16")]; - string var_9027_pad_type_0 = const()[name = string("op_9027_pad_type_0"), val = string("valid")]; - tensor var_9027_strides_0 = const()[name = string("op_9027_strides_0"), val = tensor([1, 1])]; - tensor var_9027_pad_0 = const()[name = string("op_9027_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9027_dilations_0 = const()[name = string("op_9027_dilations_0"), val = tensor([1, 1])]; - int32 var_9027_groups_0 = const()[name = string("op_9027_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306273664))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306666944))))[name = string("layers_23_self_attn_k_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_9027_cast_fp16 = conv(dilations = var_9027_dilations_0, groups = var_9027_groups_0, pad = var_9027_pad_0, pad_type = var_9027_pad_type_0, strides = var_9027_strides_0, weight = layers_23_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_95_cast_fp16)[name = string("op_9027_cast_fp16")]; - string var_9033_pad_type_0 = const()[name = string("op_9033_pad_type_0"), val = string("valid")]; - tensor var_9033_strides_0 = const()[name = string("op_9033_strides_0"), val = tensor([1, 1])]; - tensor var_9033_pad_0 = const()[name = string("op_9033_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9033_dilations_0 = const()[name = string("op_9033_dilations_0"), val = tensor([1, 1])]; - int32 var_9033_groups_0 = const()[name = string("op_9033_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306676800))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306668032))))[name = string("layers_23_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9033_cast_fp16 = conv(dilations = var_9033_dilations_0, groups = var_9033_groups_0, pad = var_9033_pad_0, pad_type = var_9033_pad_type_0, strides = var_9033_strides_0, weight = layers_23_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_95_cast_fp16)[name = string("op_9033_cast_fp16")]; - tensor key_cast_fp16 = add(x = var_9027_cast_fp16, y = var_9033_cast_fp16)[name = string("key_cast_fp16")]; - string var_9043_pad_type_0 = const()[name = string("op_9043_pad_type_0"), val = string("valid")]; - tensor var_9043_strides_0 = const()[name = string("op_9043_strides_0"), val = tensor([1, 1])]; - tensor var_9043_pad_0 = const()[name = string("op_9043_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9043_dilations_0 = const()[name = string("op_9043_dilations_0"), val = tensor([1, 1])]; - int32 var_9043_groups_0 = const()[name = string("op_9043_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306807936))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307201216))))[name = string("layers_23_self_attn_v_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_9043_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9043_dilations_0, groups = var_9043_groups_0, pad = var_9043_pad_0, pad_type = var_9043_pad_type_0, strides = var_9043_strides_0, weight = layers_23_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_95_cast_fp16)[name = string("op_9043_cast_fp16")]; - string var_9049_pad_type_0 = const()[name = string("op_9049_pad_type_0"), val = string("valid")]; - tensor var_9049_strides_0 = const()[name = string("op_9049_strides_0"), val = tensor([1, 1])]; - tensor var_9049_pad_0 = const()[name = string("op_9049_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9049_dilations_0 = const()[name = string("op_9049_dilations_0"), val = tensor([1, 1])]; - int32 var_9049_groups_0 = const()[name = string("op_9049_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307209920))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307202304))))[name = string("layers_23_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9049_cast_fp16 = conv(dilations = var_9049_dilations_0, groups = var_9049_groups_0, pad = var_9049_pad_0, pad_type = var_9049_pad_type_0, strides = var_9049_strides_0, weight = layers_23_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_95_cast_fp16)[name = string("op_9049_cast_fp16")]; - tensor value_cast_fp16 = add(x = var_9043_cast_fp16, y = var_9049_cast_fp16)[name = string("value_cast_fp16")]; - tensor var_9052_to_fp16 = const()[name = string("op_9052_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307341056)))]; - tensor query_cast_fp16 = add(x = query_93_cast_fp16, y = var_9052_to_fp16)[name = string("query_cast_fp16")]; - tensor var_9055_to_fp16 = const()[name = string("op_9055_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307343168)))]; - tensor q_with_bias_v_cast_fp16 = add(x = query_93_cast_fp16, y = var_9055_to_fp16)[name = string("q_with_bias_v_cast_fp16")]; - string var_9065_pad_type_0 = const()[name = string("op_9065_pad_type_0"), val = string("valid")]; - tensor var_9065_strides_0 = const()[name = string("op_9065_strides_0"), val = tensor([1, 1])]; - tensor var_9065_pad_0 = const()[name = string("op_9065_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9065_dilations_0 = const()[name = string("op_9065_dilations_0"), val = tensor([1, 1])]; - int32 var_9065_groups_0 = const()[name = string("op_9065_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307345280))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307738560))))[name = string("layers_23_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized")]; - tensor var_9065_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9065_dilations_0, groups = var_9065_groups_0, pad = var_9065_pad_0, pad_type = var_9065_pad_type_0, strides = var_9065_strides_0, weight = layers_23_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = pos_enc_to_fp16)[name = string("op_9065_cast_fp16")]; - string var_9071_pad_type_0 = const()[name = string("op_9071_pad_type_0"), val = string("valid")]; - tensor var_9071_strides_0 = const()[name = string("op_9071_strides_0"), val = tensor([1, 1])]; - tensor var_9071_pad_0 = const()[name = string("op_9071_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9071_dilations_0 = const()[name = string("op_9071_dilations_0"), val = tensor([1, 1])]; - int32 var_9071_groups_0 = const()[name = string("op_9071_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307761472))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307739648))))[name = string("layers_23_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9071_cast_fp16 = conv(dilations = var_9071_dilations_0, groups = var_9071_groups_0, pad = var_9071_pad_0, pad_type = var_9071_pad_type_0, strides = var_9071_strides_0, weight = layers_23_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = pos_enc_to_fp16)[name = string("op_9071_cast_fp16")]; - tensor p_cast_fp16 = add(x = var_9065_cast_fp16, y = var_9071_cast_fp16)[name = string("p_cast_fp16")]; - tensor var_9075 = const()[name = string("op_9075"), val = tensor([1, 8, 128, 188])]; - tensor var_9076_cast_fp16 = reshape(shape = var_9075, x = q_with_bias_v_cast_fp16)[name = string("op_9076_cast_fp16")]; - tensor var_9077 = const()[name = string("op_9077"), val = tensor([1, 8, 128, -1])]; - tensor var_9078_cast_fp16 = reshape(shape = var_9077, x = p_cast_fp16)[name = string("op_9078_cast_fp16")]; - bool matrix_bd_185_transpose_x_0 = const()[name = string("matrix_bd_185_transpose_x_0"), val = bool(true)]; - bool matrix_bd_185_transpose_y_0 = const()[name = string("matrix_bd_185_transpose_y_0"), val = bool(false)]; - tensor matrix_bd_185_cast_fp16 = matmul(transpose_x = matrix_bd_185_transpose_x_0, transpose_y = matrix_bd_185_transpose_y_0, x = var_9076_cast_fp16, y = var_9078_cast_fp16)[name = string("matrix_bd_185_cast_fp16")]; - tensor matrix_bd_187_pad_0 = const()[name = string("matrix_bd_187_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; - string matrix_bd_187_mode_0 = const()[name = string("matrix_bd_187_mode_0"), val = string("constant")]; - fp16 const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = fp16(0x0p+0)]; - tensor matrix_bd_187_cast_fp16 = pad(constant_val = const_263_to_fp16, mode = matrix_bd_187_mode_0, pad = matrix_bd_187_pad_0, x = matrix_bd_185_cast_fp16)[name = string("matrix_bd_187_cast_fp16")]; - tensor var_9087 = const()[name = string("op_9087"), val = tensor([1, 8, -1, 188])]; - tensor matrix_bd_189_cast_fp16 = reshape(shape = var_9087, x = matrix_bd_187_cast_fp16)[name = string("matrix_bd_189_cast_fp16")]; - tensor var_9091_begin_0 = const()[name = string("op_9091_begin_0"), val = tensor([0, 0, 1, 0])]; - tensor var_9091_end_0 = const()[name = string("op_9091_end_0"), val = tensor([1, 8, 376, 188])]; - tensor var_9091_end_mask_0 = const()[name = string("op_9091_end_mask_0"), val = tensor([true, true, true, true])]; - tensor var_9091_cast_fp16 = slice_by_index(begin = var_9091_begin_0, end = var_9091_end_0, end_mask = var_9091_end_mask_0, x = matrix_bd_189_cast_fp16)[name = string("op_9091_cast_fp16")]; - tensor var_9092 = const()[name = string("op_9092"), val = tensor([1, 8, 188, 375])]; - tensor matrix_bd_cast_fp16 = reshape(shape = var_9092, x = var_9091_cast_fp16)[name = string("matrix_bd_cast_fp16")]; - tensor var_9097_begin_0 = const()[name = string("op_9097_begin_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9097_end_0 = const()[name = string("op_9097_end_0"), val = tensor([1, 8, 188, 188])]; - tensor var_9097_end_mask_0 = const()[name = string("op_9097_end_mask_0"), val = tensor([true, true, true, false])]; - tensor var_9097_cast_fp16 = slice_by_index(begin = var_9097_begin_0, end = var_9097_end_0, end_mask = var_9097_end_mask_0, x = matrix_bd_cast_fp16)[name = string("op_9097_cast_fp16")]; - fp16 var_9098_to_fp16 = const()[name = string("op_9098_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor qk_mask_cast_fp16 = mul(x = var_9097_cast_fp16, y = var_9098_to_fp16)[name = string("qk_mask_cast_fp16")]; - tensor var_9102 = const()[name = string("op_9102"), val = tensor([1, 8, 128, 188])]; - tensor mh_q_cast_fp16 = reshape(shape = var_9102, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; - fp16 var_9104_to_fp16 = const()[name = string("op_9104_to_fp16"), val = fp16(0x1.6ap-4)]; - tensor var_9105_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_9104_to_fp16)[name = string("op_9105_cast_fp16")]; - tensor var_9108 = const()[name = string("op_9108"), val = tensor([1, 8, 128, 188])]; - tensor var_9109_cast_fp16 = reshape(shape = var_9108, x = key_cast_fp16)[name = string("op_9109_cast_fp16")]; - bool mh_w_93_transpose_x_0 = const()[name = string("mh_w_93_transpose_x_0"), val = bool(true)]; - bool mh_w_93_transpose_y_0 = const()[name = string("mh_w_93_transpose_y_0"), val = bool(false)]; - tensor mh_w_93_cast_fp16 = matmul(transpose_x = mh_w_93_transpose_x_0, transpose_y = mh_w_93_transpose_y_0, x = var_9105_cast_fp16, y = var_9109_cast_fp16)[name = string("mh_w_93_cast_fp16")]; - tensor mh_w_cast_fp16 = add(x = mh_w_93_cast_fp16, y = qk_mask_cast_fp16)[name = string("mh_w_cast_fp16")]; - tensor var_9113_cast_fp16 = softmax(axis = var_8900, x = mh_w_cast_fp16)[name = string("op_9113_cast_fp16")]; - tensor var_9114 = const()[name = string("op_9114"), val = tensor([1, 8, 128, 188])]; - tensor var_9115_cast_fp16 = reshape(shape = var_9114, x = value_cast_fp16)[name = string("op_9115_cast_fp16")]; - bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; - bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; - tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_9115_cast_fp16, y = var_9113_cast_fp16)[name = string("attn_cast_fp16")]; - tensor var_9118 = const()[name = string("op_9118"), val = tensor([1, 1024, 1, 188])]; - tensor input_621_cast_fp16 = reshape(shape = var_9118, x = attn_cast_fp16)[name = string("input_621_cast_fp16")]; - string var_9128_pad_type_0 = const()[name = string("op_9128_pad_type_0"), val = string("valid")]; - tensor var_9128_strides_0 = const()[name = string("op_9128_strides_0"), val = tensor([1, 1])]; - tensor var_9128_pad_0 = const()[name = string("op_9128_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9128_dilations_0 = const()[name = string("op_9128_dilations_0"), val = tensor([1, 1])]; - int32 var_9128_groups_0 = const()[name = string("op_9128_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307892608))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308285888))))[name = string("layers_23_self_attn_o_proj_inlier_module_weight_to_fp16_palettized")]; - tensor var_9128_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9128_dilations_0, groups = var_9128_groups_0, pad = var_9128_pad_0, pad_type = var_9128_pad_type_0, strides = var_9128_strides_0, weight = layers_23_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_621_cast_fp16)[name = string("op_9128_cast_fp16")]; - string var_9134_pad_type_0 = const()[name = string("op_9134_pad_type_0"), val = string("valid")]; - tensor var_9134_strides_0 = const()[name = string("op_9134_strides_0"), val = tensor([1, 1])]; - tensor var_9134_pad_0 = const()[name = string("op_9134_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9134_dilations_0 = const()[name = string("op_9134_dilations_0"), val = tensor([1, 1])]; - int32 var_9134_groups_0 = const()[name = string("op_9134_groups_0"), val = int32(1)]; - tensor layers_23_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308296128))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308286976))))[name = string("layers_23_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9134_cast_fp16 = conv(dilations = var_9134_dilations_0, groups = var_9134_groups_0, pad = var_9134_pad_0, pad_type = var_9134_pad_type_0, strides = var_9134_strides_0, weight = layers_23_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_621_cast_fp16)[name = string("op_9134_cast_fp16")]; - tensor obj_cast_fp16 = add(x = var_9128_cast_fp16, y = var_9134_cast_fp16)[name = string("obj_cast_fp16")]; - tensor inputs_235_cast_fp16 = add(x = inputs_233_cast_fp16, y = obj_cast_fp16)[name = string("inputs_235_cast_fp16")]; - tensor out_235_axes_0 = const()[name = string("out_235_axes_0"), val = tensor([1])]; - fp16 var_9145_to_fp16 = const()[name = string("op_9145_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_235_cast_fp16 = layer_norm(axes = out_235_axes_0, epsilon = var_9145_to_fp16, x = inputs_235_cast_fp16)[name = string("out_235_cast_fp16")]; - tensor input_623_gamma_0_to_fp16 = const()[name = string("input_623_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308427264)))]; - tensor input_623_beta_0_to_fp16 = const()[name = string("input_623_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308429376)))]; - fp16 input_623_epsilon_0_to_fp16 = const()[name = string("input_623_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_623_cast_fp16 = batch_norm(beta = input_623_beta_0_to_fp16, epsilon = input_623_epsilon_0_to_fp16, gamma = input_623_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_235_cast_fp16)[name = string("input_623_cast_fp16")]; - string var_9166_pad_type_0 = const()[name = string("op_9166_pad_type_0"), val = string("valid")]; - tensor var_9166_strides_0 = const()[name = string("op_9166_strides_0"), val = tensor([1, 1])]; - tensor var_9166_pad_0 = const()[name = string("op_9166_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9166_dilations_0 = const()[name = string("op_9166_dilations_0"), val = tensor([1, 1])]; - int32 var_9166_groups_0 = const()[name = string("op_9166_groups_0"), val = int32(1)]; - tensor layers_23_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308431488))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309217984))))[name = string("layers_23_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized")]; - tensor var_9166_cast_fp16 = conv(dilations = var_9166_dilations_0, groups = var_9166_groups_0, pad = var_9166_pad_0, pad_type = var_9166_pad_type_0, strides = var_9166_strides_0, weight = layers_23_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_623_cast_fp16)[name = string("op_9166_cast_fp16")]; - string var_9172_pad_type_0 = const()[name = string("op_9172_pad_type_0"), val = string("valid")]; - tensor var_9172_strides_0 = const()[name = string("op_9172_strides_0"), val = tensor([1, 1])]; - tensor var_9172_pad_0 = const()[name = string("op_9172_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9172_dilations_0 = const()[name = string("op_9172_dilations_0"), val = tensor([1, 1])]; - int32 var_9172_groups_0 = const()[name = string("op_9172_groups_0"), val = int32(1)]; - tensor layers_23_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309240896))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309220096))))[name = string("layers_23_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9172_cast_fp16 = conv(dilations = var_9172_dilations_0, groups = var_9172_groups_0, pad = var_9172_pad_0, pad_type = var_9172_pad_type_0, strides = var_9172_strides_0, weight = layers_23_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_623_cast_fp16)[name = string("op_9172_cast_fp16")]; - tensor input_625_cast_fp16 = add(x = var_9166_cast_fp16, y = var_9172_cast_fp16)[name = string("input_625_cast_fp16")]; - int32 input_627_split_num_splits_0 = const()[name = string("input_627_split_num_splits_0"), val = int32(2)]; - int32 input_627_split_axis_0 = const()[name = string("input_627_split_axis_0"), val = int32(1)]; - tensor input_627_split_cast_fp16_0, tensor input_627_split_cast_fp16_1 = split(axis = input_627_split_axis_0, num_splits = input_627_split_num_splits_0, x = input_625_cast_fp16)[name = string("input_627_split_cast_fp16")]; - tensor input_627_split_1_sigmoid_cast_fp16 = sigmoid(x = input_627_split_cast_fp16_1)[name = string("input_627_split_1_sigmoid_cast_fp16")]; - tensor input_627_cast_fp16 = mul(x = input_627_split_cast_fp16_0, y = input_627_split_1_sigmoid_cast_fp16)[name = string("input_627_cast_fp16")]; - string input_629_pad_type_0 = const()[name = string("input_629_pad_type_0"), val = string("custom")]; - tensor input_629_pad_0 = const()[name = string("input_629_pad_0"), val = tensor([0, 0, 4, 4])]; - int32 input_629_groups_0 = const()[name = string("input_629_groups_0"), val = int32(1024)]; - tensor input_629_strides_0 = const()[name = string("input_629_strides_0"), val = tensor([1, 1])]; - tensor input_629_dilations_0 = const()[name = string("input_629_dilations_0"), val = tensor([1, 1])]; - tensor const_314_to_fp16 = const()[name = string("const_314_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309503104)))]; - tensor const_315_to_fp16 = const()[name = string("const_315_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309521600)))]; - tensor input_631_cast_fp16 = conv(bias = const_315_to_fp16, dilations = input_629_dilations_0, groups = input_629_groups_0, pad = input_629_pad_0, pad_type = input_629_pad_type_0, strides = input_629_strides_0, weight = const_314_to_fp16, x = input_627_cast_fp16)[name = string("input_631_cast_fp16")]; - tensor input_633_cast_fp16 = silu(x = input_631_cast_fp16)[name = string("input_633_cast_fp16")]; - string var_9194_pad_type_0 = const()[name = string("op_9194_pad_type_0"), val = string("valid")]; - tensor var_9194_strides_0 = const()[name = string("op_9194_strides_0"), val = tensor([1, 1])]; - tensor var_9194_pad_0 = const()[name = string("op_9194_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9194_dilations_0 = const()[name = string("op_9194_dilations_0"), val = tensor([1, 1])]; - int32 var_9194_groups_0 = const()[name = string("op_9194_groups_0"), val = int32(1)]; - tensor layers_23_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309523712))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309916992))))[name = string("layers_23_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized")]; - tensor var_9194_cast_fp16 = conv(dilations = var_9194_dilations_0, groups = var_9194_groups_0, pad = var_9194_pad_0, pad_type = var_9194_pad_type_0, strides = var_9194_strides_0, weight = layers_23_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_633_cast_fp16)[name = string("op_9194_cast_fp16")]; - string var_9200_pad_type_0 = const()[name = string("op_9200_pad_type_0"), val = string("valid")]; - tensor var_9200_strides_0 = const()[name = string("op_9200_strides_0"), val = tensor([1, 1])]; - tensor var_9200_pad_0 = const()[name = string("op_9200_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9200_dilations_0 = const()[name = string("op_9200_dilations_0"), val = tensor([1, 1])]; - int32 var_9200_groups_0 = const()[name = string("op_9200_groups_0"), val = int32(1)]; - tensor layers_23_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309931904))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309918080))))[name = string("layers_23_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9200_cast_fp16 = conv(dilations = var_9200_dilations_0, groups = var_9200_groups_0, pad = var_9200_pad_0, pad_type = var_9200_pad_type_0, strides = var_9200_strides_0, weight = layers_23_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_633_cast_fp16)[name = string("op_9200_cast_fp16")]; - tensor x_143_cast_fp16 = add(x = var_9194_cast_fp16, y = var_9200_cast_fp16)[name = string("x_143_cast_fp16")]; - tensor inputs_237_cast_fp16 = add(x = inputs_235_cast_fp16, y = x_143_cast_fp16)[name = string("inputs_237_cast_fp16")]; - tensor out_237_axes_0 = const()[name = string("out_237_axes_0"), val = tensor([1])]; - fp16 var_9211_to_fp16 = const()[name = string("op_9211_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_237_cast_fp16 = layer_norm(axes = out_237_axes_0, epsilon = var_9211_to_fp16, x = inputs_237_cast_fp16)[name = string("out_237_cast_fp16")]; - tensor input_635_gamma_0_to_fp16 = const()[name = string("input_635_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310063040)))]; - tensor input_635_beta_0_to_fp16 = const()[name = string("input_635_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310065152)))]; - fp16 input_635_epsilon_0_to_fp16 = const()[name = string("input_635_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor input_635_cast_fp16 = batch_norm(beta = input_635_beta_0_to_fp16, epsilon = input_635_epsilon_0_to_fp16, gamma = input_635_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_237_cast_fp16)[name = string("input_635_cast_fp16")]; - string var_9231_pad_type_0 = const()[name = string("op_9231_pad_type_0"), val = string("valid")]; - tensor var_9231_strides_0 = const()[name = string("op_9231_strides_0"), val = tensor([1, 1])]; - tensor var_9231_pad_0 = const()[name = string("op_9231_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9231_dilations_0 = const()[name = string("op_9231_dilations_0"), val = tensor([1, 1])]; - int32 var_9231_groups_0 = const()[name = string("op_9231_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310067264))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311640192))))[name = string("layers_23_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized")]; - tensor var_9231_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_9231_dilations_0, groups = var_9231_groups_0, pad = var_9231_pad_0, pad_type = var_9231_pad_type_0, strides = var_9231_strides_0, weight = layers_23_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_635_cast_fp16)[name = string("op_9231_cast_fp16")]; - string var_9237_pad_type_0 = const()[name = string("op_9237_pad_type_0"), val = string("valid")]; - tensor var_9237_strides_0 = const()[name = string("op_9237_strides_0"), val = tensor([1, 1])]; - tensor var_9237_pad_0 = const()[name = string("op_9237_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9237_dilations_0 = const()[name = string("op_9237_dilations_0"), val = tensor([1, 1])]; - int32 var_9237_groups_0 = const()[name = string("op_9237_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311807040))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311644352))))[name = string("layers_23_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9237_cast_fp16 = conv(dilations = var_9237_dilations_0, groups = var_9237_groups_0, pad = var_9237_pad_0, pad_type = var_9237_pad_type_0, strides = var_9237_strides_0, weight = layers_23_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_635_cast_fp16)[name = string("op_9237_cast_fp16")]; - tensor input_637_cast_fp16 = add(x = var_9231_cast_fp16, y = var_9237_cast_fp16)[name = string("input_637_cast_fp16")]; - tensor input_639_cast_fp16 = silu(x = input_637_cast_fp16)[name = string("input_639_cast_fp16")]; - string var_9248_pad_type_0 = const()[name = string("op_9248_pad_type_0"), val = string("valid")]; - tensor var_9248_strides_0 = const()[name = string("op_9248_strides_0"), val = tensor([1, 1])]; - tensor var_9248_pad_0 = const()[name = string("op_9248_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9248_dilations_0 = const()[name = string("op_9248_dilations_0"), val = tensor([1, 1])]; - int32 var_9248_groups_0 = const()[name = string("op_9248_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312331392))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313904320))))[name = string("layers_23_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized")]; - tensor var_9248_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9248_dilations_0, groups = var_9248_groups_0, pad = var_9248_pad_0, pad_type = var_9248_pad_type_0, strides = var_9248_strides_0, weight = layers_23_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_639_cast_fp16)[name = string("op_9248_cast_fp16")]; - string var_9254_pad_type_0 = const()[name = string("op_9254_pad_type_0"), val = string("valid")]; - tensor var_9254_strides_0 = const()[name = string("op_9254_strides_0"), val = tensor([1, 1])]; - tensor var_9254_pad_0 = const()[name = string("op_9254_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9254_dilations_0 = const()[name = string("op_9254_dilations_0"), val = tensor([1, 1])]; - int32 var_9254_groups_0 = const()[name = string("op_9254_groups_0"), val = int32(1)]; - tensor layers_23_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314012864))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313905408))))[name = string("layers_23_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9254_cast_fp16 = conv(dilations = var_9254_dilations_0, groups = var_9254_groups_0, pad = var_9254_pad_0, pad_type = var_9254_pad_type_0, strides = var_9254_strides_0, weight = layers_23_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_639_cast_fp16)[name = string("op_9254_cast_fp16")]; - tensor x_cast_fp16 = add(x = var_9248_cast_fp16, y = var_9254_cast_fp16)[name = string("x_cast_fp16")]; - fp16 var_9256_to_fp16 = const()[name = string("op_9256_to_fp16"), val = fp16(0x1p-1)]; - tensor var_9257_cast_fp16 = mul(x = x_cast_fp16, y = var_9256_to_fp16)[name = string("op_9257_cast_fp16")]; - tensor inputs_cast_fp16 = add(x = inputs_237_cast_fp16, y = var_9257_cast_fp16)[name = string("inputs_cast_fp16")]; - tensor out_239_axes_0 = const()[name = string("out_239_axes_0"), val = tensor([1])]; - fp16 var_9267_to_fp16 = const()[name = string("op_9267_to_fp16"), val = fp16(0x1.5p-17)]; - tensor out_239_cast_fp16 = layer_norm(axes = out_239_axes_0, epsilon = var_9267_to_fp16, x = inputs_cast_fp16)[name = string("out_239_cast_fp16")]; - tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314537216)))]; - tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314539328)))]; - fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; - tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_239_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; - string var_9290_pad_type_0 = const()[name = string("op_9290_pad_type_0"), val = string("valid")]; - tensor var_9290_strides_0 = const()[name = string("op_9290_strides_0"), val = tensor([1, 1])]; - tensor var_9290_pad_0 = const()[name = string("op_9290_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9290_dilations_0 = const()[name = string("op_9290_dilations_0"), val = tensor([1, 1])]; - int32 var_9290_groups_0 = const()[name = string("op_9290_groups_0"), val = int32(1)]; - tensor joint_projection_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314541440))), lut = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314787264))))[name = string("joint_projection_inlier_module_weight_to_fp16_palettized")]; - tensor joint_projection_inlier_module_bias_to_fp16 = const()[name = string("joint_projection_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314787968)))]; - tensor var_9290_cast_fp16 = conv(bias = joint_projection_inlier_module_bias_to_fp16, dilations = var_9290_dilations_0, groups = var_9290_groups_0, pad = var_9290_pad_0, pad_type = var_9290_pad_type_0, strides = var_9290_strides_0, weight = joint_projection_inlier_module_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("op_9290_cast_fp16")]; - string var_9296_pad_type_0 = const()[name = string("op_9296_pad_type_0"), val = string("valid")]; - tensor var_9296_strides_0 = const()[name = string("op_9296_strides_0"), val = tensor([1, 1])]; - tensor var_9296_pad_0 = const()[name = string("op_9296_pad_0"), val = tensor([0, 0, 0, 0])]; - tensor var_9296_dilations_0 = const()[name = string("op_9296_dilations_0"), val = tensor([1, 1])]; - int32 var_9296_groups_0 = const()[name = string("op_9296_groups_0"), val = int32(1)]; - tensor joint_projection_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense(mask = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314795328))), nonzero_data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314789312))))[name = string("joint_projection_outlier_module_weight_to_fp16_sparsified")]; - tensor var_9296_cast_fp16 = conv(dilations = var_9296_dilations_0, groups = var_9296_groups_0, pad = var_9296_pad_0, pad_type = var_9296_pad_type_0, strides = var_9296_strides_0, weight = joint_projection_outlier_module_weight_to_fp16_sparsified, x = encoder_output_embeds)[name = string("op_9296_cast_fp16")]; - tensor joint_projected_encoder_output_embeds = add(x = var_9290_cast_fp16, y = var_9296_cast_fp16)[name = string("op_9297_cast_fp16")]; - } -> (encoder_output_embeds, joint_projected_encoder_output_embeds); -} \ No newline at end of file