program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.2"}})]
{
    func main<ios16>(tensor<fp16, [1, 80, 1, 3000]> melspectrogram_features) {
            tensor<string, []> var_90_pad_type_0 = const()[name = tensor<string, []>("op_90_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_90_pad_0 = const()[name = tensor<string, []>("op_90_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])];
            tensor<int32, [2]> var_90_strides_0 = const()[name = tensor<string, []>("op_90_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_90_dilations_0 = const()[name = tensor<string, []>("op_90_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_90_groups_0 = const()[name = tensor<string, []>("op_90_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 80, 1, 3]> var_65_to_fp16 = const()[name = tensor<string, []>("op_65_to_fp16"), val = tensor<fp16, [1024, 80, 1, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp16, [1024]> var_71_to_fp16 = const()[name = tensor<string, []>("op_71_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491648)))];
            tensor<fp16, [1, 1024, 1, 3000]> var_90_cast_fp16 = conv(bias = var_71_to_fp16, dilations = var_90_dilations_0, groups = var_90_groups_0, pad = var_90_pad_0, pad_type = var_90_pad_type_0, strides = var_90_strides_0, weight = var_65_to_fp16, x = melspectrogram_features)[name = tensor<string, []>("op_90_cast_fp16")];
            tensor<string, []> hidden_states_1_mode_0 = const()[name = tensor<string, []>("hidden_states_1_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 1, 3000]> hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_90_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")];
            tensor<string, []> var_130_pad_type_0 = const()[name = tensor<string, []>("op_130_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_130_pad_0 = const()[name = tensor<string, []>("op_130_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])];
            tensor<int32, [2]> var_130_strides_0 = const()[name = tensor<string, []>("op_130_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_130_dilations_0 = const()[name = tensor<string, []>("op_130_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_130_groups_0 = const()[name = tensor<string, []>("op_130_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 3]> var_105_to_fp16 = const()[name = tensor<string, []>("op_105_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493760)))];
            tensor<fp16, [1024]> var_111_to_fp16 = const()[name = tensor<string, []>("op_111_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6785280)))];
            tensor<fp16, [1, 1024, 1, 1500]> var_130_cast_fp16 = conv(bias = var_111_to_fp16, dilations = var_130_dilations_0, groups = var_130_groups_0, pad = var_130_pad_0, pad_type = var_130_pad_type_0, strides = var_130_strides_0, weight = var_105_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("op_130_cast_fp16")];
            tensor<string, []> hidden_states_3_mode_0 = const()[name = tensor<string, []>("hidden_states_3_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_130_cast_fp16)[name = tensor<string, []>("hidden_states_3_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> var_148_to_fp16 = const()[name = tensor<string, []>("op_148_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1500]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6787392)))];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_148_to_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")];
            tensor<int32, []> var_158 = const()[name = tensor<string, []>("op_158"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_1_axes_0 = const()[name = tensor<string, []>("out_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_180_to_fp16 = const()[name = tensor<string, []>("op_180_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_180_to_fp16, x = inputs_1_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")];
            tensor<fp16, [1024]> obj_1_mean_0_to_fp16 = const()[name = tensor<string, []>("obj_1_mean_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9859456)))];
            tensor<fp16, [1024]> obj_1_variance_0_to_fp16 = const()[name = tensor<string, []>("obj_1_variance_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9861568)))];
            tensor<fp16, [1024]> obj_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9863680)))];
            tensor<fp16, [1024]> obj_1_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_1_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9865792)))];
            tensor<fp16, []> obj_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor<string, []>("obj_1_cast_fp16")];
            tensor<string, []> query_1_pad_type_0 = const()[name = tensor<string, []>("query_1_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_1_strides_0 = const()[name = tensor<string, []>("query_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_1_pad_0 = const()[name = tensor<string, []>("query_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_1_dilations_0 = const()[name = tensor<string, []>("query_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_1_groups_0 = const()[name = tensor<string, []>("query_1_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9867904)))];
            tensor<fp16, [1024]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11965120)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("query_1_cast_fp16")];
            tensor<string, []> key_1_pad_type_0 = const()[name = tensor<string, []>("key_1_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_1_strides_0 = const()[name = tensor<string, []>("key_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_1_pad_0 = const()[name = tensor<string, []>("key_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_1_dilations_0 = const()[name = tensor<string, []>("key_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_1_groups_0 = const()[name = tensor<string, []>("key_1_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11967232)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")];
            tensor<string, []> value_1_pad_type_0 = const()[name = tensor<string, []>("value_1_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_1_strides_0 = const()[name = tensor<string, []>("value_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_1_pad_0 = const()[name = tensor<string, []>("value_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_1_dilations_0 = const()[name = tensor<string, []>("value_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_1_groups_0 = const()[name = tensor<string, []>("value_1_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14064448)))];
            tensor<fp16, [1024]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16161664)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")];
            tensor<int32, [4]> var_216 = const()[name = tensor<string, []>("op_216"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_1_cast_fp16 = reshape(shape = var_216, x = query_1_cast_fp16)[name = tensor<string, []>("mh_q_1_cast_fp16")];
            tensor<fp16, []> var_218_to_fp16 = const()[name = tensor<string, []>("op_218_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_219_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_218_to_fp16)[name = tensor<string, []>("op_219_cast_fp16")];
            tensor<int32, [4]> var_222 = const()[name = tensor<string, []>("op_222"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_223_cast_fp16 = reshape(shape = var_222, x = key_1_cast_fp16)[name = tensor<string, []>("op_223_cast_fp16")];
            tensor<bool, []> mh_w_1_transpose_x_0 = const()[name = tensor<string, []>("mh_w_1_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_1_transpose_y_0 = const()[name = tensor<string, []>("mh_w_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_219_cast_fp16, y = var_223_cast_fp16)[name = tensor<string, []>("mh_w_1_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_226_cast_fp16 = softmax(axis = var_158, x = mh_w_1_cast_fp16)[name = tensor<string, []>("op_226_cast_fp16")];
            tensor<int32, [4]> var_227 = const()[name = tensor<string, []>("op_227"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_228_cast_fp16 = reshape(shape = var_227, x = value_1_cast_fp16)[name = tensor<string, []>("op_228_cast_fp16")];
            tensor<bool, []> attn_1_transpose_x_0 = const()[name = tensor<string, []>("attn_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_1_transpose_y_0 = const()[name = tensor<string, []>("attn_1_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_228_cast_fp16, y = var_226_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")];
            tensor<int32, [4]> var_231 = const()[name = tensor<string, []>("op_231"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_1_cast_fp16 = reshape(shape = var_231, x = attn_1_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")];
            tensor<string, []> obj_3_pad_type_0 = const()[name = tensor<string, []>("obj_3_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_3_strides_0 = const()[name = tensor<string, []>("obj_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_3_pad_0 = const()[name = tensor<string, []>("obj_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_3_dilations_0 = const()[name = tensor<string, []>("obj_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_3_groups_0 = const()[name = tensor<string, []>("obj_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16163776)))];
            tensor<fp16, [1024]> layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18260992)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("obj_3_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = tensor<string, []>("inputs_3_cast_fp16")];
            tensor<int32, [1]> out_3_axes_0 = const()[name = tensor<string, []>("out_3_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_249_to_fp16 = const()[name = tensor<string, []>("op_249_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_249_to_fp16, x = inputs_3_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")];
            tensor<fp16, [1024]> input_3_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_3_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18263104)))];
            tensor<fp16, [1024]> input_3_beta_0_to_fp16 = const()[name = tensor<string, []>("input_3_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18265216)))];
            tensor<fp16, []> input_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
            tensor<string, []> input_5_pad_type_0 = const()[name = tensor<string, []>("input_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_5_strides_0 = const()[name = tensor<string, []>("input_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_5_pad_0 = const()[name = tensor<string, []>("input_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_5_dilations_0 = const()[name = tensor<string, []>("input_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_5_groups_0 = const()[name = tensor<string, []>("input_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18267328)))];
            tensor<fp16, [4096]> layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26656000)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_5_cast_fp16 = conv(bias = layers_0_fc1_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 = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<string, []> input_7_mode_0 = const()[name = tensor<string, []>("input_7_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<string, []> hidden_states_5_pad_type_0 = const()[name = tensor<string, []>("hidden_states_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_5_strides_0 = const()[name = tensor<string, []>("hidden_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = tensor<string, []>("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_5_dilations_0 = const()[name = tensor<string, []>("hidden_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_5_groups_0 = const()[name = tensor<string, []>("hidden_states_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26664256)))];
            tensor<fp16, [1024]> layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35052928)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("hidden_states_5_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor<string, []>("inputs_5_cast_fp16")];
            tensor<int32, []> var_278 = const()[name = tensor<string, []>("op_278"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_5_axes_0 = const()[name = tensor<string, []>("out_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_300_to_fp16 = const()[name = tensor<string, []>("op_300_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_300_to_fp16, x = inputs_5_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")];
            tensor<fp16, [1024]> obj_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_5_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35055040)))];
            tensor<fp16, [1024]> obj_5_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_5_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35057152)))];
            tensor<fp16, []> obj_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor<string, []>("obj_5_cast_fp16")];
            tensor<string, []> query_3_pad_type_0 = const()[name = tensor<string, []>("query_3_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_3_strides_0 = const()[name = tensor<string, []>("query_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_3_pad_0 = const()[name = tensor<string, []>("query_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_3_dilations_0 = const()[name = tensor<string, []>("query_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_3_groups_0 = const()[name = tensor<string, []>("query_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35059264)))];
            tensor<fp16, [1024]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37156480)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor<string, []>("query_3_cast_fp16")];
            tensor<string, []> key_3_pad_type_0 = const()[name = tensor<string, []>("key_3_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_3_strides_0 = const()[name = tensor<string, []>("key_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_3_pad_0 = const()[name = tensor<string, []>("key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_3_dilations_0 = const()[name = tensor<string, []>("key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_3_groups_0 = const()[name = tensor<string, []>("key_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37158592)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor<string, []>("key_3_cast_fp16")];
            tensor<string, []> value_3_pad_type_0 = const()[name = tensor<string, []>("value_3_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_3_strides_0 = const()[name = tensor<string, []>("value_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_3_pad_0 = const()[name = tensor<string, []>("value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_3_dilations_0 = const()[name = tensor<string, []>("value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_3_groups_0 = const()[name = tensor<string, []>("value_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39255808)))];
            tensor<fp16, [1024]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41353024)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor<string, []>("value_3_cast_fp16")];
            tensor<int32, [4]> var_336 = const()[name = tensor<string, []>("op_336"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_3_cast_fp16 = reshape(shape = var_336, x = query_3_cast_fp16)[name = tensor<string, []>("mh_q_3_cast_fp16")];
            tensor<fp16, []> var_338_to_fp16 = const()[name = tensor<string, []>("op_338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_339_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_338_to_fp16)[name = tensor<string, []>("op_339_cast_fp16")];
            tensor<int32, [4]> var_342 = const()[name = tensor<string, []>("op_342"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_343_cast_fp16 = reshape(shape = var_342, x = key_3_cast_fp16)[name = tensor<string, []>("op_343_cast_fp16")];
            tensor<bool, []> mh_w_3_transpose_x_0 = const()[name = tensor<string, []>("mh_w_3_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_3_transpose_y_0 = const()[name = tensor<string, []>("mh_w_3_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_3_cast_fp16 = matmul(transpose_x = mh_w_3_transpose_x_0, transpose_y = mh_w_3_transpose_y_0, x = var_339_cast_fp16, y = var_343_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_346_cast_fp16 = softmax(axis = var_278, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_346_cast_fp16")];
            tensor<int32, [4]> var_347 = const()[name = tensor<string, []>("op_347"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_348_cast_fp16 = reshape(shape = var_347, x = value_3_cast_fp16)[name = tensor<string, []>("op_348_cast_fp16")];
            tensor<bool, []> attn_3_transpose_x_0 = const()[name = tensor<string, []>("attn_3_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_3_transpose_y_0 = const()[name = tensor<string, []>("attn_3_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_348_cast_fp16, y = var_346_cast_fp16)[name = tensor<string, []>("attn_3_cast_fp16")];
            tensor<int32, [4]> var_351 = const()[name = tensor<string, []>("op_351"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_9_cast_fp16 = reshape(shape = var_351, x = attn_3_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
            tensor<string, []> obj_7_pad_type_0 = const()[name = tensor<string, []>("obj_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_7_strides_0 = const()[name = tensor<string, []>("obj_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_7_pad_0 = const()[name = tensor<string, []>("obj_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_7_dilations_0 = const()[name = tensor<string, []>("obj_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_7_groups_0 = const()[name = tensor<string, []>("obj_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41355136)))];
            tensor<fp16, [1024]> layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43452352)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("obj_7_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = tensor<string, []>("inputs_7_cast_fp16")];
            tensor<int32, [1]> out_7_axes_0 = const()[name = tensor<string, []>("out_7_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_369_to_fp16 = const()[name = tensor<string, []>("op_369_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_369_to_fp16, x = inputs_7_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")];
            tensor<fp16, [1024]> input_11_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_11_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43454464)))];
            tensor<fp16, [1024]> input_11_beta_0_to_fp16 = const()[name = tensor<string, []>("input_11_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43456576)))];
            tensor<fp16, []> input_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
            tensor<string, []> input_13_pad_type_0 = const()[name = tensor<string, []>("input_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_13_strides_0 = const()[name = tensor<string, []>("input_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_13_pad_0 = const()[name = tensor<string, []>("input_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_13_dilations_0 = const()[name = tensor<string, []>("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_13_groups_0 = const()[name = tensor<string, []>("input_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_1_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43458688)))];
            tensor<fp16, [4096]> layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51847360)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_13_cast_fp16 = conv(bias = layers_1_fc1_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 = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<string, []> input_15_mode_0 = const()[name = tensor<string, []>("input_15_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
            tensor<string, []> hidden_states_7_pad_type_0 = const()[name = tensor<string, []>("hidden_states_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = tensor<string, []>("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = tensor<string, []>("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = tensor<string, []>("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_7_groups_0 = const()[name = tensor<string, []>("hidden_states_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_1_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51855616)))];
            tensor<fp16, [1024]> layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60244288)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor<string, []>("inputs_9_cast_fp16")];
            tensor<int32, []> var_398 = const()[name = tensor<string, []>("op_398"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_9_axes_0 = const()[name = tensor<string, []>("out_9_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_420_to_fp16 = const()[name = tensor<string, []>("op_420_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_420_to_fp16, x = inputs_9_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")];
            tensor<fp16, [1024]> obj_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60246400)))];
            tensor<fp16, [1024]> obj_9_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_9_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60248512)))];
            tensor<fp16, []> obj_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor<string, []>("obj_9_cast_fp16")];
            tensor<string, []> query_5_pad_type_0 = const()[name = tensor<string, []>("query_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_5_strides_0 = const()[name = tensor<string, []>("query_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_5_pad_0 = const()[name = tensor<string, []>("query_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_5_dilations_0 = const()[name = tensor<string, []>("query_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_5_groups_0 = const()[name = tensor<string, []>("query_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60250624)))];
            tensor<fp16, [1024]> layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62347840)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor<string, []>("query_5_cast_fp16")];
            tensor<string, []> key_5_pad_type_0 = const()[name = tensor<string, []>("key_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_5_strides_0 = const()[name = tensor<string, []>("key_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_5_pad_0 = const()[name = tensor<string, []>("key_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_5_dilations_0 = const()[name = tensor<string, []>("key_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_5_groups_0 = const()[name = tensor<string, []>("key_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62349952)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")];
            tensor<string, []> value_5_pad_type_0 = const()[name = tensor<string, []>("value_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_5_strides_0 = const()[name = tensor<string, []>("value_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_5_pad_0 = const()[name = tensor<string, []>("value_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_5_dilations_0 = const()[name = tensor<string, []>("value_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_5_groups_0 = const()[name = tensor<string, []>("value_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64447168)))];
            tensor<fp16, [1024]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66544384)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")];
            tensor<int32, [4]> var_456 = const()[name = tensor<string, []>("op_456"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_5_cast_fp16 = reshape(shape = var_456, x = query_5_cast_fp16)[name = tensor<string, []>("mh_q_5_cast_fp16")];
            tensor<fp16, []> var_458_to_fp16 = const()[name = tensor<string, []>("op_458_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_459_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_458_to_fp16)[name = tensor<string, []>("op_459_cast_fp16")];
            tensor<int32, [4]> var_462 = const()[name = tensor<string, []>("op_462"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_463_cast_fp16 = reshape(shape = var_462, x = key_5_cast_fp16)[name = tensor<string, []>("op_463_cast_fp16")];
            tensor<bool, []> mh_w_5_transpose_x_0 = const()[name = tensor<string, []>("mh_w_5_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_5_transpose_y_0 = const()[name = tensor<string, []>("mh_w_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_459_cast_fp16, y = var_463_cast_fp16)[name = tensor<string, []>("mh_w_5_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_466_cast_fp16 = softmax(axis = var_398, x = mh_w_5_cast_fp16)[name = tensor<string, []>("op_466_cast_fp16")];
            tensor<int32, [4]> var_467 = const()[name = tensor<string, []>("op_467"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_468_cast_fp16 = reshape(shape = var_467, x = value_5_cast_fp16)[name = tensor<string, []>("op_468_cast_fp16")];
            tensor<bool, []> attn_5_transpose_x_0 = const()[name = tensor<string, []>("attn_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_5_transpose_y_0 = const()[name = tensor<string, []>("attn_5_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_468_cast_fp16, y = var_466_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")];
            tensor<int32, [4]> var_471 = const()[name = tensor<string, []>("op_471"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_17_cast_fp16 = reshape(shape = var_471, x = attn_5_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<string, []> obj_11_pad_type_0 = const()[name = tensor<string, []>("obj_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_11_strides_0 = const()[name = tensor<string, []>("obj_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_11_pad_0 = const()[name = tensor<string, []>("obj_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_11_dilations_0 = const()[name = tensor<string, []>("obj_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_11_groups_0 = const()[name = tensor<string, []>("obj_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66546496)))];
            tensor<fp16, [1024]> layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68643712)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("obj_11_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = tensor<string, []>("inputs_11_cast_fp16")];
            tensor<int32, [1]> out_11_axes_0 = const()[name = tensor<string, []>("out_11_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_489_to_fp16 = const()[name = tensor<string, []>("op_489_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_489_to_fp16, x = inputs_11_cast_fp16)[name = tensor<string, []>("out_11_cast_fp16")];
            tensor<fp16, [1024]> input_19_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_19_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68645824)))];
            tensor<fp16, [1024]> input_19_beta_0_to_fp16 = const()[name = tensor<string, []>("input_19_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68647936)))];
            tensor<fp16, []> input_19_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_19_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
            tensor<string, []> input_21_pad_type_0 = const()[name = tensor<string, []>("input_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_21_strides_0 = const()[name = tensor<string, []>("input_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_21_pad_0 = const()[name = tensor<string, []>("input_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_21_dilations_0 = const()[name = tensor<string, []>("input_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_21_groups_0 = const()[name = tensor<string, []>("input_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_2_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68650048)))];
            tensor<fp16, [4096]> layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77038720)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
            tensor<string, []> input_23_mode_0 = const()[name = tensor<string, []>("input_23_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")];
            tensor<string, []> hidden_states_9_pad_type_0 = const()[name = tensor<string, []>("hidden_states_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_9_strides_0 = const()[name = tensor<string, []>("hidden_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_9_pad_0 = const()[name = tensor<string, []>("hidden_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_9_dilations_0 = const()[name = tensor<string, []>("hidden_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_9_groups_0 = const()[name = tensor<string, []>("hidden_states_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_2_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77046976)))];
            tensor<fp16, [1024]> layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85435648)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("hidden_states_9_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = tensor<string, []>("inputs_13_cast_fp16")];
            tensor<int32, []> var_518 = const()[name = tensor<string, []>("op_518"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_13_axes_0 = const()[name = tensor<string, []>("out_13_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_540_to_fp16 = const()[name = tensor<string, []>("op_540_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_540_to_fp16, x = inputs_13_cast_fp16)[name = tensor<string, []>("out_13_cast_fp16")];
            tensor<fp16, [1024]> obj_13_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_13_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85437760)))];
            tensor<fp16, [1024]> obj_13_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_13_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85439872)))];
            tensor<fp16, []> obj_13_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_13_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor<string, []>("obj_13_cast_fp16")];
            tensor<string, []> query_7_pad_type_0 = const()[name = tensor<string, []>("query_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_7_strides_0 = const()[name = tensor<string, []>("query_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_7_pad_0 = const()[name = tensor<string, []>("query_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_7_dilations_0 = const()[name = tensor<string, []>("query_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_7_groups_0 = const()[name = tensor<string, []>("query_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85441984)))];
            tensor<fp16, [1024]> layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87539200)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor<string, []>("query_7_cast_fp16")];
            tensor<string, []> key_7_pad_type_0 = const()[name = tensor<string, []>("key_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_7_strides_0 = const()[name = tensor<string, []>("key_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_7_pad_0 = const()[name = tensor<string, []>("key_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_7_dilations_0 = const()[name = tensor<string, []>("key_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_7_groups_0 = const()[name = tensor<string, []>("key_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87541312)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor<string, []>("key_7_cast_fp16")];
            tensor<string, []> value_7_pad_type_0 = const()[name = tensor<string, []>("value_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_7_strides_0 = const()[name = tensor<string, []>("value_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_7_pad_0 = const()[name = tensor<string, []>("value_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_7_dilations_0 = const()[name = tensor<string, []>("value_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_7_groups_0 = const()[name = tensor<string, []>("value_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89638528)))];
            tensor<fp16, [1024]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91735744)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor<string, []>("value_7_cast_fp16")];
            tensor<int32, [4]> var_576 = const()[name = tensor<string, []>("op_576"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_7_cast_fp16 = reshape(shape = var_576, x = query_7_cast_fp16)[name = tensor<string, []>("mh_q_7_cast_fp16")];
            tensor<fp16, []> var_578_to_fp16 = const()[name = tensor<string, []>("op_578_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_579_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_578_to_fp16)[name = tensor<string, []>("op_579_cast_fp16")];
            tensor<int32, [4]> var_582 = const()[name = tensor<string, []>("op_582"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_583_cast_fp16 = reshape(shape = var_582, x = key_7_cast_fp16)[name = tensor<string, []>("op_583_cast_fp16")];
            tensor<bool, []> mh_w_7_transpose_x_0 = const()[name = tensor<string, []>("mh_w_7_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_7_transpose_y_0 = const()[name = tensor<string, []>("mh_w_7_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_579_cast_fp16, y = var_583_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_586_cast_fp16 = softmax(axis = var_518, x = mh_w_7_cast_fp16)[name = tensor<string, []>("op_586_cast_fp16")];
            tensor<int32, [4]> var_587 = const()[name = tensor<string, []>("op_587"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_588_cast_fp16 = reshape(shape = var_587, x = value_7_cast_fp16)[name = tensor<string, []>("op_588_cast_fp16")];
            tensor<bool, []> attn_7_transpose_x_0 = const()[name = tensor<string, []>("attn_7_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_7_transpose_y_0 = const()[name = tensor<string, []>("attn_7_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_588_cast_fp16, y = var_586_cast_fp16)[name = tensor<string, []>("attn_7_cast_fp16")];
            tensor<int32, [4]> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_25_cast_fp16 = reshape(shape = var_591, x = attn_7_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<string, []> obj_15_pad_type_0 = const()[name = tensor<string, []>("obj_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_15_strides_0 = const()[name = tensor<string, []>("obj_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_15_pad_0 = const()[name = tensor<string, []>("obj_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_15_dilations_0 = const()[name = tensor<string, []>("obj_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_15_groups_0 = const()[name = tensor<string, []>("obj_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91737856)))];
            tensor<fp16, [1024]> layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93835072)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("obj_15_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = tensor<string, []>("inputs_15_cast_fp16")];
            tensor<int32, [1]> out_15_axes_0 = const()[name = tensor<string, []>("out_15_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_609_to_fp16 = const()[name = tensor<string, []>("op_609_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_609_to_fp16, x = inputs_15_cast_fp16)[name = tensor<string, []>("out_15_cast_fp16")];
            tensor<fp16, [1024]> input_27_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_27_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93837184)))];
            tensor<fp16, [1024]> input_27_beta_0_to_fp16 = const()[name = tensor<string, []>("input_27_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93839296)))];
            tensor<fp16, []> input_27_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_27_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")];
            tensor<string, []> input_29_pad_type_0 = const()[name = tensor<string, []>("input_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_29_strides_0 = const()[name = tensor<string, []>("input_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_29_pad_0 = const()[name = tensor<string, []>("input_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_29_dilations_0 = const()[name = tensor<string, []>("input_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_29_groups_0 = const()[name = tensor<string, []>("input_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_3_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93841408)))];
            tensor<fp16, [4096]> layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102230080)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")];
            tensor<string, []> input_31_mode_0 = const()[name = tensor<string, []>("input_31_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<string, []> hidden_states_11_pad_type_0 = const()[name = tensor<string, []>("hidden_states_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_11_strides_0 = const()[name = tensor<string, []>("hidden_states_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_11_pad_0 = const()[name = tensor<string, []>("hidden_states_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_11_dilations_0 = const()[name = tensor<string, []>("hidden_states_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_11_groups_0 = const()[name = tensor<string, []>("hidden_states_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_3_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102238336)))];
            tensor<fp16, [1024]> layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110627008)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("hidden_states_11_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor<string, []>("inputs_17_cast_fp16")];
            tensor<int32, []> var_638 = const()[name = tensor<string, []>("op_638"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_17_axes_0 = const()[name = tensor<string, []>("out_17_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_660_to_fp16 = const()[name = tensor<string, []>("op_660_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_660_to_fp16, x = inputs_17_cast_fp16)[name = tensor<string, []>("out_17_cast_fp16")];
            tensor<fp16, [1024]> obj_17_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_17_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110629120)))];
            tensor<fp16, [1024]> obj_17_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_17_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110631232)))];
            tensor<fp16, []> obj_17_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_17_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor<string, []>("obj_17_cast_fp16")];
            tensor<string, []> query_9_pad_type_0 = const()[name = tensor<string, []>("query_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_9_strides_0 = const()[name = tensor<string, []>("query_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_9_pad_0 = const()[name = tensor<string, []>("query_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_9_dilations_0 = const()[name = tensor<string, []>("query_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_9_groups_0 = const()[name = tensor<string, []>("query_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110633344)))];
            tensor<fp16, [1024]> layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112730560)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor<string, []>("query_9_cast_fp16")];
            tensor<string, []> key_9_pad_type_0 = const()[name = tensor<string, []>("key_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_9_strides_0 = const()[name = tensor<string, []>("key_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_9_pad_0 = const()[name = tensor<string, []>("key_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_9_dilations_0 = const()[name = tensor<string, []>("key_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_9_groups_0 = const()[name = tensor<string, []>("key_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112732672)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor<string, []>("key_9_cast_fp16")];
            tensor<string, []> value_9_pad_type_0 = const()[name = tensor<string, []>("value_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_9_strides_0 = const()[name = tensor<string, []>("value_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_9_pad_0 = const()[name = tensor<string, []>("value_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_9_dilations_0 = const()[name = tensor<string, []>("value_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_9_groups_0 = const()[name = tensor<string, []>("value_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114829888)))];
            tensor<fp16, [1024]> layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116927104)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor<string, []>("value_9_cast_fp16")];
            tensor<int32, [4]> var_696 = const()[name = tensor<string, []>("op_696"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_9_cast_fp16 = reshape(shape = var_696, x = query_9_cast_fp16)[name = tensor<string, []>("mh_q_9_cast_fp16")];
            tensor<fp16, []> var_698_to_fp16 = const()[name = tensor<string, []>("op_698_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_699_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_698_to_fp16)[name = tensor<string, []>("op_699_cast_fp16")];
            tensor<int32, [4]> var_702 = const()[name = tensor<string, []>("op_702"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_703_cast_fp16 = reshape(shape = var_702, x = key_9_cast_fp16)[name = tensor<string, []>("op_703_cast_fp16")];
            tensor<bool, []> mh_w_9_transpose_x_0 = const()[name = tensor<string, []>("mh_w_9_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_9_transpose_y_0 = const()[name = tensor<string, []>("mh_w_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_699_cast_fp16, y = var_703_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_706_cast_fp16 = softmax(axis = var_638, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_706_cast_fp16")];
            tensor<int32, [4]> var_707 = const()[name = tensor<string, []>("op_707"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_708_cast_fp16 = reshape(shape = var_707, x = value_9_cast_fp16)[name = tensor<string, []>("op_708_cast_fp16")];
            tensor<bool, []> attn_9_transpose_x_0 = const()[name = tensor<string, []>("attn_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_9_transpose_y_0 = const()[name = tensor<string, []>("attn_9_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_708_cast_fp16, y = var_706_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")];
            tensor<int32, [4]> var_711 = const()[name = tensor<string, []>("op_711"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_33_cast_fp16 = reshape(shape = var_711, x = attn_9_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")];
            tensor<string, []> obj_19_pad_type_0 = const()[name = tensor<string, []>("obj_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_19_strides_0 = const()[name = tensor<string, []>("obj_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_19_pad_0 = const()[name = tensor<string, []>("obj_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_19_dilations_0 = const()[name = tensor<string, []>("obj_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_19_groups_0 = const()[name = tensor<string, []>("obj_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116929216)))];
            tensor<fp16, [1024]> layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119026432)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("obj_19_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = tensor<string, []>("inputs_19_cast_fp16")];
            tensor<int32, [1]> out_19_axes_0 = const()[name = tensor<string, []>("out_19_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_729_to_fp16 = const()[name = tensor<string, []>("op_729_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_729_to_fp16, x = inputs_19_cast_fp16)[name = tensor<string, []>("out_19_cast_fp16")];
            tensor<fp16, [1024]> input_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_35_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119028544)))];
            tensor<fp16, [1024]> input_35_beta_0_to_fp16 = const()[name = tensor<string, []>("input_35_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119030656)))];
            tensor<fp16, []> input_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<string, []> input_37_pad_type_0 = const()[name = tensor<string, []>("input_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_37_strides_0 = const()[name = tensor<string, []>("input_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_37_pad_0 = const()[name = tensor<string, []>("input_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_37_dilations_0 = const()[name = tensor<string, []>("input_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_37_groups_0 = const()[name = tensor<string, []>("input_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_4_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119032768)))];
            tensor<fp16, [4096]> layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127421440)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<string, []> input_39_mode_0 = const()[name = tensor<string, []>("input_39_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")];
            tensor<string, []> hidden_states_13_pad_type_0 = const()[name = tensor<string, []>("hidden_states_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_13_strides_0 = const()[name = tensor<string, []>("hidden_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_13_pad_0 = const()[name = tensor<string, []>("hidden_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_13_dilations_0 = const()[name = tensor<string, []>("hidden_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_13_groups_0 = const()[name = tensor<string, []>("hidden_states_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_4_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127429696)))];
            tensor<fp16, [1024]> layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135818368)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor<string, []>("inputs_21_cast_fp16")];
            tensor<int32, []> var_758 = const()[name = tensor<string, []>("op_758"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_21_axes_0 = const()[name = tensor<string, []>("out_21_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_780_to_fp16 = const()[name = tensor<string, []>("op_780_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_780_to_fp16, x = inputs_21_cast_fp16)[name = tensor<string, []>("out_21_cast_fp16")];
            tensor<fp16, [1024]> obj_21_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_21_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135820480)))];
            tensor<fp16, [1024]> obj_21_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_21_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135822592)))];
            tensor<fp16, []> obj_21_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_21_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor<string, []>("obj_21_cast_fp16")];
            tensor<string, []> query_11_pad_type_0 = const()[name = tensor<string, []>("query_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_11_strides_0 = const()[name = tensor<string, []>("query_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_11_pad_0 = const()[name = tensor<string, []>("query_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_11_dilations_0 = const()[name = tensor<string, []>("query_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_11_groups_0 = const()[name = tensor<string, []>("query_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135824704)))];
            tensor<fp16, [1024]> layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137921920)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor<string, []>("query_11_cast_fp16")];
            tensor<string, []> key_11_pad_type_0 = const()[name = tensor<string, []>("key_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_11_strides_0 = const()[name = tensor<string, []>("key_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_11_pad_0 = const()[name = tensor<string, []>("key_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_11_dilations_0 = const()[name = tensor<string, []>("key_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_11_groups_0 = const()[name = tensor<string, []>("key_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137924032)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor<string, []>("key_11_cast_fp16")];
            tensor<string, []> value_11_pad_type_0 = const()[name = tensor<string, []>("value_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_11_strides_0 = const()[name = tensor<string, []>("value_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_11_pad_0 = const()[name = tensor<string, []>("value_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_11_dilations_0 = const()[name = tensor<string, []>("value_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_11_groups_0 = const()[name = tensor<string, []>("value_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140021248)))];
            tensor<fp16, [1024]> layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142118464)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor<string, []>("value_11_cast_fp16")];
            tensor<int32, [4]> var_816 = const()[name = tensor<string, []>("op_816"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_11_cast_fp16 = reshape(shape = var_816, x = query_11_cast_fp16)[name = tensor<string, []>("mh_q_11_cast_fp16")];
            tensor<fp16, []> var_818_to_fp16 = const()[name = tensor<string, []>("op_818_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_819_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_818_to_fp16)[name = tensor<string, []>("op_819_cast_fp16")];
            tensor<int32, [4]> var_822 = const()[name = tensor<string, []>("op_822"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_823_cast_fp16 = reshape(shape = var_822, x = key_11_cast_fp16)[name = tensor<string, []>("op_823_cast_fp16")];
            tensor<bool, []> mh_w_11_transpose_x_0 = const()[name = tensor<string, []>("mh_w_11_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_11_transpose_y_0 = const()[name = tensor<string, []>("mh_w_11_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_819_cast_fp16, y = var_823_cast_fp16)[name = tensor<string, []>("mh_w_11_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_826_cast_fp16 = softmax(axis = var_758, x = mh_w_11_cast_fp16)[name = tensor<string, []>("op_826_cast_fp16")];
            tensor<int32, [4]> var_827 = const()[name = tensor<string, []>("op_827"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_828_cast_fp16 = reshape(shape = var_827, x = value_11_cast_fp16)[name = tensor<string, []>("op_828_cast_fp16")];
            tensor<bool, []> attn_11_transpose_x_0 = const()[name = tensor<string, []>("attn_11_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_11_transpose_y_0 = const()[name = tensor<string, []>("attn_11_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_828_cast_fp16, y = var_826_cast_fp16)[name = tensor<string, []>("attn_11_cast_fp16")];
            tensor<int32, [4]> var_831 = const()[name = tensor<string, []>("op_831"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_41_cast_fp16 = reshape(shape = var_831, x = attn_11_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
            tensor<string, []> obj_23_pad_type_0 = const()[name = tensor<string, []>("obj_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_23_strides_0 = const()[name = tensor<string, []>("obj_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_23_pad_0 = const()[name = tensor<string, []>("obj_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_23_dilations_0 = const()[name = tensor<string, []>("obj_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_23_groups_0 = const()[name = tensor<string, []>("obj_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142120576)))];
            tensor<fp16, [1024]> layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144217792)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("obj_23_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = tensor<string, []>("inputs_23_cast_fp16")];
            tensor<int32, [1]> out_23_axes_0 = const()[name = tensor<string, []>("out_23_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_849_to_fp16 = const()[name = tensor<string, []>("op_849_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_849_to_fp16, x = inputs_23_cast_fp16)[name = tensor<string, []>("out_23_cast_fp16")];
            tensor<fp16, [1024]> input_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_43_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144219904)))];
            tensor<fp16, [1024]> input_43_beta_0_to_fp16 = const()[name = tensor<string, []>("input_43_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144222016)))];
            tensor<fp16, []> input_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
            tensor<string, []> input_45_pad_type_0 = const()[name = tensor<string, []>("input_45_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_45_strides_0 = const()[name = tensor<string, []>("input_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_45_pad_0 = const()[name = tensor<string, []>("input_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_45_dilations_0 = const()[name = tensor<string, []>("input_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_45_groups_0 = const()[name = tensor<string, []>("input_45_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_5_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144224128)))];
            tensor<fp16, [4096]> layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152612800)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")];
            tensor<string, []> input_47_mode_0 = const()[name = tensor<string, []>("input_47_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<string, []> hidden_states_15_pad_type_0 = const()[name = tensor<string, []>("hidden_states_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_15_strides_0 = const()[name = tensor<string, []>("hidden_states_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_15_pad_0 = const()[name = tensor<string, []>("hidden_states_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_15_dilations_0 = const()[name = tensor<string, []>("hidden_states_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_15_groups_0 = const()[name = tensor<string, []>("hidden_states_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_5_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152621056)))];
            tensor<fp16, [1024]> layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161009728)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = tensor<string, []>("hidden_states_15_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = tensor<string, []>("inputs_25_cast_fp16")];
            tensor<int32, []> var_878 = const()[name = tensor<string, []>("op_878"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_25_axes_0 = const()[name = tensor<string, []>("out_25_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_900_to_fp16 = const()[name = tensor<string, []>("op_900_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_900_to_fp16, x = inputs_25_cast_fp16)[name = tensor<string, []>("out_25_cast_fp16")];
            tensor<fp16, [1024]> obj_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_25_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161011840)))];
            tensor<fp16, [1024]> obj_25_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_25_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161013952)))];
            tensor<fp16, []> obj_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor<string, []>("obj_25_cast_fp16")];
            tensor<string, []> query_13_pad_type_0 = const()[name = tensor<string, []>("query_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_13_strides_0 = const()[name = tensor<string, []>("query_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_13_pad_0 = const()[name = tensor<string, []>("query_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_13_dilations_0 = const()[name = tensor<string, []>("query_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_13_groups_0 = const()[name = tensor<string, []>("query_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161016064)))];
            tensor<fp16, [1024]> layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163113280)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor<string, []>("query_13_cast_fp16")];
            tensor<string, []> key_13_pad_type_0 = const()[name = tensor<string, []>("key_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_13_strides_0 = const()[name = tensor<string, []>("key_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_13_pad_0 = const()[name = tensor<string, []>("key_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_13_dilations_0 = const()[name = tensor<string, []>("key_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_13_groups_0 = const()[name = tensor<string, []>("key_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163115392)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor<string, []>("key_13_cast_fp16")];
            tensor<string, []> value_13_pad_type_0 = const()[name = tensor<string, []>("value_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_13_strides_0 = const()[name = tensor<string, []>("value_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_13_pad_0 = const()[name = tensor<string, []>("value_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_13_dilations_0 = const()[name = tensor<string, []>("value_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_13_groups_0 = const()[name = tensor<string, []>("value_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165212608)))];
            tensor<fp16, [1024]> layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167309824)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor<string, []>("value_13_cast_fp16")];
            tensor<int32, [4]> var_936 = const()[name = tensor<string, []>("op_936"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_13_cast_fp16 = reshape(shape = var_936, x = query_13_cast_fp16)[name = tensor<string, []>("mh_q_13_cast_fp16")];
            tensor<fp16, []> var_938_to_fp16 = const()[name = tensor<string, []>("op_938_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_939_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_938_to_fp16)[name = tensor<string, []>("op_939_cast_fp16")];
            tensor<int32, [4]> var_942 = const()[name = tensor<string, []>("op_942"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_943_cast_fp16 = reshape(shape = var_942, x = key_13_cast_fp16)[name = tensor<string, []>("op_943_cast_fp16")];
            tensor<bool, []> mh_w_13_transpose_x_0 = const()[name = tensor<string, []>("mh_w_13_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_13_transpose_y_0 = const()[name = tensor<string, []>("mh_w_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_939_cast_fp16, y = var_943_cast_fp16)[name = tensor<string, []>("mh_w_13_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_946_cast_fp16 = softmax(axis = var_878, x = mh_w_13_cast_fp16)[name = tensor<string, []>("op_946_cast_fp16")];
            tensor<int32, [4]> var_947 = const()[name = tensor<string, []>("op_947"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_948_cast_fp16 = reshape(shape = var_947, x = value_13_cast_fp16)[name = tensor<string, []>("op_948_cast_fp16")];
            tensor<bool, []> attn_13_transpose_x_0 = const()[name = tensor<string, []>("attn_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_13_transpose_y_0 = const()[name = tensor<string, []>("attn_13_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_948_cast_fp16, y = var_946_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")];
            tensor<int32, [4]> var_951 = const()[name = tensor<string, []>("op_951"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_49_cast_fp16 = reshape(shape = var_951, x = attn_13_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
            tensor<string, []> obj_27_pad_type_0 = const()[name = tensor<string, []>("obj_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_27_strides_0 = const()[name = tensor<string, []>("obj_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_27_pad_0 = const()[name = tensor<string, []>("obj_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_27_dilations_0 = const()[name = tensor<string, []>("obj_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_27_groups_0 = const()[name = tensor<string, []>("obj_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167311936)))];
            tensor<fp16, [1024]> layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169409152)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("obj_27_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = tensor<string, []>("inputs_27_cast_fp16")];
            tensor<int32, [1]> out_27_axes_0 = const()[name = tensor<string, []>("out_27_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_969_to_fp16 = const()[name = tensor<string, []>("op_969_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_969_to_fp16, x = inputs_27_cast_fp16)[name = tensor<string, []>("out_27_cast_fp16")];
            tensor<fp16, [1024]> input_51_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_51_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169411264)))];
            tensor<fp16, [1024]> input_51_beta_0_to_fp16 = const()[name = tensor<string, []>("input_51_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169413376)))];
            tensor<fp16, []> input_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
            tensor<string, []> input_53_pad_type_0 = const()[name = tensor<string, []>("input_53_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_53_strides_0 = const()[name = tensor<string, []>("input_53_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_53_pad_0 = const()[name = tensor<string, []>("input_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_53_dilations_0 = const()[name = tensor<string, []>("input_53_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_53_groups_0 = const()[name = tensor<string, []>("input_53_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_6_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169415488)))];
            tensor<fp16, [4096]> layers_6_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177804160)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
            tensor<string, []> input_55_mode_0 = const()[name = tensor<string, []>("input_55_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<string, []> hidden_states_17_pad_type_0 = const()[name = tensor<string, []>("hidden_states_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_17_strides_0 = const()[name = tensor<string, []>("hidden_states_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_17_pad_0 = const()[name = tensor<string, []>("hidden_states_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_17_dilations_0 = const()[name = tensor<string, []>("hidden_states_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_17_groups_0 = const()[name = tensor<string, []>("hidden_states_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_6_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177812416)))];
            tensor<fp16, [1024]> layers_6_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186201088)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("hidden_states_17_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = tensor<string, []>("inputs_29_cast_fp16")];
            tensor<int32, []> var_998 = const()[name = tensor<string, []>("op_998"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_29_axes_0 = const()[name = tensor<string, []>("out_29_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1020_to_fp16 = const()[name = tensor<string, []>("op_1020_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1020_to_fp16, x = inputs_29_cast_fp16)[name = tensor<string, []>("out_29_cast_fp16")];
            tensor<fp16, [1024]> obj_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_29_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186203200)))];
            tensor<fp16, [1024]> obj_29_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_29_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186205312)))];
            tensor<fp16, []> obj_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor<string, []>("obj_29_cast_fp16")];
            tensor<string, []> query_15_pad_type_0 = const()[name = tensor<string, []>("query_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_15_strides_0 = const()[name = tensor<string, []>("query_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_15_pad_0 = const()[name = tensor<string, []>("query_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_15_dilations_0 = const()[name = tensor<string, []>("query_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_15_groups_0 = const()[name = tensor<string, []>("query_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186207424)))];
            tensor<fp16, [1024]> layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188304640)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor<string, []>("query_15_cast_fp16")];
            tensor<string, []> key_15_pad_type_0 = const()[name = tensor<string, []>("key_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_15_strides_0 = const()[name = tensor<string, []>("key_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_15_pad_0 = const()[name = tensor<string, []>("key_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_15_dilations_0 = const()[name = tensor<string, []>("key_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_15_groups_0 = const()[name = tensor<string, []>("key_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188306752)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor<string, []>("key_15_cast_fp16")];
            tensor<string, []> value_15_pad_type_0 = const()[name = tensor<string, []>("value_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_15_strides_0 = const()[name = tensor<string, []>("value_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_15_pad_0 = const()[name = tensor<string, []>("value_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_15_dilations_0 = const()[name = tensor<string, []>("value_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_15_groups_0 = const()[name = tensor<string, []>("value_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190403968)))];
            tensor<fp16, [1024]> layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192501184)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor<string, []>("value_15_cast_fp16")];
            tensor<int32, [4]> var_1056 = const()[name = tensor<string, []>("op_1056"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_15_cast_fp16 = reshape(shape = var_1056, x = query_15_cast_fp16)[name = tensor<string, []>("mh_q_15_cast_fp16")];
            tensor<fp16, []> var_1058_to_fp16 = const()[name = tensor<string, []>("op_1058_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1059_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_1058_to_fp16)[name = tensor<string, []>("op_1059_cast_fp16")];
            tensor<int32, [4]> var_1062 = const()[name = tensor<string, []>("op_1062"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1063_cast_fp16 = reshape(shape = var_1062, x = key_15_cast_fp16)[name = tensor<string, []>("op_1063_cast_fp16")];
            tensor<bool, []> mh_w_15_transpose_x_0 = const()[name = tensor<string, []>("mh_w_15_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_15_transpose_y_0 = const()[name = tensor<string, []>("mh_w_15_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_15_cast_fp16 = matmul(transpose_x = mh_w_15_transpose_x_0, transpose_y = mh_w_15_transpose_y_0, x = var_1059_cast_fp16, y = var_1063_cast_fp16)[name = tensor<string, []>("mh_w_15_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1066_cast_fp16 = softmax(axis = var_998, x = mh_w_15_cast_fp16)[name = tensor<string, []>("op_1066_cast_fp16")];
            tensor<int32, [4]> var_1067 = const()[name = tensor<string, []>("op_1067"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1068_cast_fp16 = reshape(shape = var_1067, x = value_15_cast_fp16)[name = tensor<string, []>("op_1068_cast_fp16")];
            tensor<bool, []> attn_15_transpose_x_0 = const()[name = tensor<string, []>("attn_15_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_15_transpose_y_0 = const()[name = tensor<string, []>("attn_15_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_1068_cast_fp16, y = var_1066_cast_fp16)[name = tensor<string, []>("attn_15_cast_fp16")];
            tensor<int32, [4]> var_1071 = const()[name = tensor<string, []>("op_1071"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_57_cast_fp16 = reshape(shape = var_1071, x = attn_15_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")];
            tensor<string, []> obj_31_pad_type_0 = const()[name = tensor<string, []>("obj_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_31_strides_0 = const()[name = tensor<string, []>("obj_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_31_pad_0 = const()[name = tensor<string, []>("obj_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_31_dilations_0 = const()[name = tensor<string, []>("obj_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_31_groups_0 = const()[name = tensor<string, []>("obj_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192503296)))];
            tensor<fp16, [1024]> layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194600512)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("obj_31_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = tensor<string, []>("inputs_31_cast_fp16")];
            tensor<int32, [1]> out_31_axes_0 = const()[name = tensor<string, []>("out_31_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1089_to_fp16 = const()[name = tensor<string, []>("op_1089_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1089_to_fp16, x = inputs_31_cast_fp16)[name = tensor<string, []>("out_31_cast_fp16")];
            tensor<fp16, [1024]> input_59_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_59_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194602624)))];
            tensor<fp16, [1024]> input_59_beta_0_to_fp16 = const()[name = tensor<string, []>("input_59_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194604736)))];
            tensor<fp16, []> input_59_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_59_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
            tensor<string, []> input_61_pad_type_0 = const()[name = tensor<string, []>("input_61_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_61_strides_0 = const()[name = tensor<string, []>("input_61_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_61_pad_0 = const()[name = tensor<string, []>("input_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_61_dilations_0 = const()[name = tensor<string, []>("input_61_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_61_groups_0 = const()[name = tensor<string, []>("input_61_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_7_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194606848)))];
            tensor<fp16, [4096]> layers_7_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202995520)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<string, []> input_63_mode_0 = const()[name = tensor<string, []>("input_63_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")];
            tensor<string, []> hidden_states_19_pad_type_0 = const()[name = tensor<string, []>("hidden_states_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_19_strides_0 = const()[name = tensor<string, []>("hidden_states_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_19_pad_0 = const()[name = tensor<string, []>("hidden_states_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_19_dilations_0 = const()[name = tensor<string, []>("hidden_states_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_19_groups_0 = const()[name = tensor<string, []>("hidden_states_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_7_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203003776)))];
            tensor<fp16, [1024]> layers_7_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211392448)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = tensor<string, []>("inputs_33_cast_fp16")];
            tensor<int32, []> var_1118 = const()[name = tensor<string, []>("op_1118"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_33_axes_0 = const()[name = tensor<string, []>("out_33_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1140_to_fp16 = const()[name = tensor<string, []>("op_1140_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1140_to_fp16, x = inputs_33_cast_fp16)[name = tensor<string, []>("out_33_cast_fp16")];
            tensor<fp16, [1024]> obj_33_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_33_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211394560)))];
            tensor<fp16, [1024]> obj_33_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_33_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211396672)))];
            tensor<fp16, []> obj_33_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_33_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor<string, []>("obj_33_cast_fp16")];
            tensor<string, []> query_17_pad_type_0 = const()[name = tensor<string, []>("query_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_17_strides_0 = const()[name = tensor<string, []>("query_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_17_pad_0 = const()[name = tensor<string, []>("query_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_17_dilations_0 = const()[name = tensor<string, []>("query_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_17_groups_0 = const()[name = tensor<string, []>("query_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211398784)))];
            tensor<fp16, [1024]> layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213496000)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor<string, []>("query_17_cast_fp16")];
            tensor<string, []> key_17_pad_type_0 = const()[name = tensor<string, []>("key_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_17_strides_0 = const()[name = tensor<string, []>("key_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_17_pad_0 = const()[name = tensor<string, []>("key_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_17_dilations_0 = const()[name = tensor<string, []>("key_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_17_groups_0 = const()[name = tensor<string, []>("key_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213498112)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor<string, []>("key_17_cast_fp16")];
            tensor<string, []> value_17_pad_type_0 = const()[name = tensor<string, []>("value_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_17_strides_0 = const()[name = tensor<string, []>("value_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_17_pad_0 = const()[name = tensor<string, []>("value_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_17_dilations_0 = const()[name = tensor<string, []>("value_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_17_groups_0 = const()[name = tensor<string, []>("value_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215595328)))];
            tensor<fp16, [1024]> layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217692544)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor<string, []>("value_17_cast_fp16")];
            tensor<int32, [4]> var_1176 = const()[name = tensor<string, []>("op_1176"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_17_cast_fp16 = reshape(shape = var_1176, x = query_17_cast_fp16)[name = tensor<string, []>("mh_q_17_cast_fp16")];
            tensor<fp16, []> var_1178_to_fp16 = const()[name = tensor<string, []>("op_1178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1179_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1178_to_fp16)[name = tensor<string, []>("op_1179_cast_fp16")];
            tensor<int32, [4]> var_1182 = const()[name = tensor<string, []>("op_1182"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1183_cast_fp16 = reshape(shape = var_1182, x = key_17_cast_fp16)[name = tensor<string, []>("op_1183_cast_fp16")];
            tensor<bool, []> mh_w_17_transpose_x_0 = const()[name = tensor<string, []>("mh_w_17_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_17_transpose_y_0 = const()[name = tensor<string, []>("mh_w_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_1179_cast_fp16, y = var_1183_cast_fp16)[name = tensor<string, []>("mh_w_17_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1186_cast_fp16 = softmax(axis = var_1118, x = mh_w_17_cast_fp16)[name = tensor<string, []>("op_1186_cast_fp16")];
            tensor<int32, [4]> var_1187 = const()[name = tensor<string, []>("op_1187"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1188_cast_fp16 = reshape(shape = var_1187, x = value_17_cast_fp16)[name = tensor<string, []>("op_1188_cast_fp16")];
            tensor<bool, []> attn_17_transpose_x_0 = const()[name = tensor<string, []>("attn_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_17_transpose_y_0 = const()[name = tensor<string, []>("attn_17_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1188_cast_fp16, y = var_1186_cast_fp16)[name = tensor<string, []>("attn_17_cast_fp16")];
            tensor<int32, [4]> var_1191 = const()[name = tensor<string, []>("op_1191"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_65_cast_fp16 = reshape(shape = var_1191, x = attn_17_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
            tensor<string, []> obj_35_pad_type_0 = const()[name = tensor<string, []>("obj_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_35_strides_0 = const()[name = tensor<string, []>("obj_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_35_pad_0 = const()[name = tensor<string, []>("obj_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_35_dilations_0 = const()[name = tensor<string, []>("obj_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_35_groups_0 = const()[name = tensor<string, []>("obj_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217694656)))];
            tensor<fp16, [1024]> layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219791872)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("obj_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = tensor<string, []>("inputs_35_cast_fp16")];
            tensor<int32, [1]> out_35_axes_0 = const()[name = tensor<string, []>("out_35_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1209_to_fp16 = const()[name = tensor<string, []>("op_1209_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1209_to_fp16, x = inputs_35_cast_fp16)[name = tensor<string, []>("out_35_cast_fp16")];
            tensor<fp16, [1024]> input_67_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_67_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219793984)))];
            tensor<fp16, [1024]> input_67_beta_0_to_fp16 = const()[name = tensor<string, []>("input_67_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219796096)))];
            tensor<fp16, []> input_67_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_67_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")];
            tensor<string, []> input_69_pad_type_0 = const()[name = tensor<string, []>("input_69_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_69_strides_0 = const()[name = tensor<string, []>("input_69_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_69_pad_0 = const()[name = tensor<string, []>("input_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_69_dilations_0 = const()[name = tensor<string, []>("input_69_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_69_groups_0 = const()[name = tensor<string, []>("input_69_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_8_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219798208)))];
            tensor<fp16, [4096]> layers_8_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228186880)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
            tensor<string, []> input_71_mode_0 = const()[name = tensor<string, []>("input_71_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
            tensor<string, []> hidden_states_21_pad_type_0 = const()[name = tensor<string, []>("hidden_states_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_21_strides_0 = const()[name = tensor<string, []>("hidden_states_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_21_pad_0 = const()[name = tensor<string, []>("hidden_states_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_21_dilations_0 = const()[name = tensor<string, []>("hidden_states_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_21_groups_0 = const()[name = tensor<string, []>("hidden_states_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_8_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228195136)))];
            tensor<fp16, [1024]> layers_8_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236583808)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = tensor<string, []>("hidden_states_21_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = tensor<string, []>("inputs_37_cast_fp16")];
            tensor<int32, []> var_1238 = const()[name = tensor<string, []>("op_1238"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_37_axes_0 = const()[name = tensor<string, []>("out_37_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1260_to_fp16 = const()[name = tensor<string, []>("op_1260_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1260_to_fp16, x = inputs_37_cast_fp16)[name = tensor<string, []>("out_37_cast_fp16")];
            tensor<fp16, [1024]> obj_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_37_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236585920)))];
            tensor<fp16, [1024]> obj_37_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_37_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236588032)))];
            tensor<fp16, []> obj_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor<string, []>("obj_37_cast_fp16")];
            tensor<string, []> query_19_pad_type_0 = const()[name = tensor<string, []>("query_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_19_strides_0 = const()[name = tensor<string, []>("query_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_19_pad_0 = const()[name = tensor<string, []>("query_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_19_dilations_0 = const()[name = tensor<string, []>("query_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_19_groups_0 = const()[name = tensor<string, []>("query_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236590144)))];
            tensor<fp16, [1024]> layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238687360)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor<string, []>("query_19_cast_fp16")];
            tensor<string, []> key_19_pad_type_0 = const()[name = tensor<string, []>("key_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_19_strides_0 = const()[name = tensor<string, []>("key_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_19_pad_0 = const()[name = tensor<string, []>("key_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_19_dilations_0 = const()[name = tensor<string, []>("key_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_19_groups_0 = const()[name = tensor<string, []>("key_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238689472)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor<string, []>("key_19_cast_fp16")];
            tensor<string, []> value_19_pad_type_0 = const()[name = tensor<string, []>("value_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_19_strides_0 = const()[name = tensor<string, []>("value_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_19_pad_0 = const()[name = tensor<string, []>("value_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_19_dilations_0 = const()[name = tensor<string, []>("value_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_19_groups_0 = const()[name = tensor<string, []>("value_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(240786688)))];
            tensor<fp16, [1024]> layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(242883904)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor<string, []>("value_19_cast_fp16")];
            tensor<int32, [4]> var_1296 = const()[name = tensor<string, []>("op_1296"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_19_cast_fp16 = reshape(shape = var_1296, x = query_19_cast_fp16)[name = tensor<string, []>("mh_q_19_cast_fp16")];
            tensor<fp16, []> var_1298_to_fp16 = const()[name = tensor<string, []>("op_1298_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1299_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1298_to_fp16)[name = tensor<string, []>("op_1299_cast_fp16")];
            tensor<int32, [4]> var_1302 = const()[name = tensor<string, []>("op_1302"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1303_cast_fp16 = reshape(shape = var_1302, x = key_19_cast_fp16)[name = tensor<string, []>("op_1303_cast_fp16")];
            tensor<bool, []> mh_w_19_transpose_x_0 = const()[name = tensor<string, []>("mh_w_19_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_19_transpose_y_0 = const()[name = tensor<string, []>("mh_w_19_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_1299_cast_fp16, y = var_1303_cast_fp16)[name = tensor<string, []>("mh_w_19_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1306_cast_fp16 = softmax(axis = var_1238, x = mh_w_19_cast_fp16)[name = tensor<string, []>("op_1306_cast_fp16")];
            tensor<int32, [4]> var_1307 = const()[name = tensor<string, []>("op_1307"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1308_cast_fp16 = reshape(shape = var_1307, x = value_19_cast_fp16)[name = tensor<string, []>("op_1308_cast_fp16")];
            tensor<bool, []> attn_19_transpose_x_0 = const()[name = tensor<string, []>("attn_19_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_19_transpose_y_0 = const()[name = tensor<string, []>("attn_19_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1308_cast_fp16, y = var_1306_cast_fp16)[name = tensor<string, []>("attn_19_cast_fp16")];
            tensor<int32, [4]> var_1311 = const()[name = tensor<string, []>("op_1311"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_73_cast_fp16 = reshape(shape = var_1311, x = attn_19_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<string, []> obj_39_pad_type_0 = const()[name = tensor<string, []>("obj_39_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_39_strides_0 = const()[name = tensor<string, []>("obj_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_39_pad_0 = const()[name = tensor<string, []>("obj_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_39_dilations_0 = const()[name = tensor<string, []>("obj_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_39_groups_0 = const()[name = tensor<string, []>("obj_39_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(242886016)))];
            tensor<fp16, [1024]> layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244983232)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("obj_39_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = tensor<string, []>("inputs_39_cast_fp16")];
            tensor<int32, [1]> out_39_axes_0 = const()[name = tensor<string, []>("out_39_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1329_to_fp16 = const()[name = tensor<string, []>("op_1329_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1329_to_fp16, x = inputs_39_cast_fp16)[name = tensor<string, []>("out_39_cast_fp16")];
            tensor<fp16, [1024]> input_75_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_75_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244985344)))];
            tensor<fp16, [1024]> input_75_beta_0_to_fp16 = const()[name = tensor<string, []>("input_75_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244987456)))];
            tensor<fp16, []> input_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")];
            tensor<string, []> input_77_pad_type_0 = const()[name = tensor<string, []>("input_77_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_77_strides_0 = const()[name = tensor<string, []>("input_77_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_77_pad_0 = const()[name = tensor<string, []>("input_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_77_dilations_0 = const()[name = tensor<string, []>("input_77_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_77_groups_0 = const()[name = tensor<string, []>("input_77_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_9_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244989568)))];
            tensor<fp16, [4096]> layers_9_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253378240)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")];
            tensor<string, []> input_79_mode_0 = const()[name = tensor<string, []>("input_79_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")];
            tensor<string, []> hidden_states_23_pad_type_0 = const()[name = tensor<string, []>("hidden_states_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_23_strides_0 = const()[name = tensor<string, []>("hidden_states_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_23_pad_0 = const()[name = tensor<string, []>("hidden_states_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_23_dilations_0 = const()[name = tensor<string, []>("hidden_states_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_23_groups_0 = const()[name = tensor<string, []>("hidden_states_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_9_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253386496)))];
            tensor<fp16, [1024]> layers_9_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261775168)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = tensor<string, []>("hidden_states_23_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor<string, []>("inputs_41_cast_fp16")];
            tensor<int32, []> var_1358 = const()[name = tensor<string, []>("op_1358"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_41_axes_0 = const()[name = tensor<string, []>("out_41_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1380_to_fp16 = const()[name = tensor<string, []>("op_1380_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1380_to_fp16, x = inputs_41_cast_fp16)[name = tensor<string, []>("out_41_cast_fp16")];
            tensor<fp16, [1024]> obj_41_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_41_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261777280)))];
            tensor<fp16, [1024]> obj_41_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_41_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261779392)))];
            tensor<fp16, []> obj_41_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_41_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor<string, []>("obj_41_cast_fp16")];
            tensor<string, []> query_21_pad_type_0 = const()[name = tensor<string, []>("query_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_21_strides_0 = const()[name = tensor<string, []>("query_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_21_pad_0 = const()[name = tensor<string, []>("query_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_21_dilations_0 = const()[name = tensor<string, []>("query_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_21_groups_0 = const()[name = tensor<string, []>("query_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261781504)))];
            tensor<fp16, [1024]> layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263878720)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor<string, []>("query_21_cast_fp16")];
            tensor<string, []> key_21_pad_type_0 = const()[name = tensor<string, []>("key_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_21_strides_0 = const()[name = tensor<string, []>("key_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_21_pad_0 = const()[name = tensor<string, []>("key_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_21_dilations_0 = const()[name = tensor<string, []>("key_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_21_groups_0 = const()[name = tensor<string, []>("key_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263880832)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor<string, []>("key_21_cast_fp16")];
            tensor<string, []> value_21_pad_type_0 = const()[name = tensor<string, []>("value_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_21_strides_0 = const()[name = tensor<string, []>("value_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_21_pad_0 = const()[name = tensor<string, []>("value_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_21_dilations_0 = const()[name = tensor<string, []>("value_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_21_groups_0 = const()[name = tensor<string, []>("value_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265978048)))];
            tensor<fp16, [1024]> layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268075264)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor<string, []>("value_21_cast_fp16")];
            tensor<int32, [4]> var_1416 = const()[name = tensor<string, []>("op_1416"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_21_cast_fp16 = reshape(shape = var_1416, x = query_21_cast_fp16)[name = tensor<string, []>("mh_q_21_cast_fp16")];
            tensor<fp16, []> var_1418_to_fp16 = const()[name = tensor<string, []>("op_1418_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1419_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1418_to_fp16)[name = tensor<string, []>("op_1419_cast_fp16")];
            tensor<int32, [4]> var_1422 = const()[name = tensor<string, []>("op_1422"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1423_cast_fp16 = reshape(shape = var_1422, x = key_21_cast_fp16)[name = tensor<string, []>("op_1423_cast_fp16")];
            tensor<bool, []> mh_w_21_transpose_x_0 = const()[name = tensor<string, []>("mh_w_21_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_21_transpose_y_0 = const()[name = tensor<string, []>("mh_w_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_1419_cast_fp16, y = var_1423_cast_fp16)[name = tensor<string, []>("mh_w_21_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1426_cast_fp16 = softmax(axis = var_1358, x = mh_w_21_cast_fp16)[name = tensor<string, []>("op_1426_cast_fp16")];
            tensor<int32, [4]> var_1427 = const()[name = tensor<string, []>("op_1427"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1428_cast_fp16 = reshape(shape = var_1427, x = value_21_cast_fp16)[name = tensor<string, []>("op_1428_cast_fp16")];
            tensor<bool, []> attn_21_transpose_x_0 = const()[name = tensor<string, []>("attn_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_21_transpose_y_0 = const()[name = tensor<string, []>("attn_21_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1428_cast_fp16, y = var_1426_cast_fp16)[name = tensor<string, []>("attn_21_cast_fp16")];
            tensor<int32, [4]> var_1431 = const()[name = tensor<string, []>("op_1431"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_81_cast_fp16 = reshape(shape = var_1431, x = attn_21_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")];
            tensor<string, []> obj_43_pad_type_0 = const()[name = tensor<string, []>("obj_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_43_strides_0 = const()[name = tensor<string, []>("obj_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_43_pad_0 = const()[name = tensor<string, []>("obj_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_43_dilations_0 = const()[name = tensor<string, []>("obj_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_43_groups_0 = const()[name = tensor<string, []>("obj_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268077376)))];
            tensor<fp16, [1024]> layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270174592)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("obj_43_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = tensor<string, []>("inputs_43_cast_fp16")];
            tensor<int32, [1]> out_43_axes_0 = const()[name = tensor<string, []>("out_43_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1449_to_fp16 = const()[name = tensor<string, []>("op_1449_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1449_to_fp16, x = inputs_43_cast_fp16)[name = tensor<string, []>("out_43_cast_fp16")];
            tensor<fp16, [1024]> input_83_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_83_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270176704)))];
            tensor<fp16, [1024]> input_83_beta_0_to_fp16 = const()[name = tensor<string, []>("input_83_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270178816)))];
            tensor<fp16, []> input_83_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_83_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")];
            tensor<string, []> input_85_pad_type_0 = const()[name = tensor<string, []>("input_85_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_85_strides_0 = const()[name = tensor<string, []>("input_85_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_85_pad_0 = const()[name = tensor<string, []>("input_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_85_dilations_0 = const()[name = tensor<string, []>("input_85_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_85_groups_0 = const()[name = tensor<string, []>("input_85_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_10_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270180928)))];
            tensor<fp16, [4096]> layers_10_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278569600)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<string, []> input_87_mode_0 = const()[name = tensor<string, []>("input_87_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")];
            tensor<string, []> hidden_states_25_pad_type_0 = const()[name = tensor<string, []>("hidden_states_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_25_strides_0 = const()[name = tensor<string, []>("hidden_states_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_25_pad_0 = const()[name = tensor<string, []>("hidden_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_25_dilations_0 = const()[name = tensor<string, []>("hidden_states_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_25_groups_0 = const()[name = tensor<string, []>("hidden_states_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_10_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278577856)))];
            tensor<fp16, [1024]> layers_10_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286966528)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = tensor<string, []>("inputs_45_cast_fp16")];
            tensor<int32, []> var_1478 = const()[name = tensor<string, []>("op_1478"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_45_axes_0 = const()[name = tensor<string, []>("out_45_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1500_to_fp16 = const()[name = tensor<string, []>("op_1500_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1500_to_fp16, x = inputs_45_cast_fp16)[name = tensor<string, []>("out_45_cast_fp16")];
            tensor<fp16, [1024]> obj_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_45_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286968640)))];
            tensor<fp16, [1024]> obj_45_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_45_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286970752)))];
            tensor<fp16, []> obj_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor<string, []>("obj_45_cast_fp16")];
            tensor<string, []> query_23_pad_type_0 = const()[name = tensor<string, []>("query_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_23_strides_0 = const()[name = tensor<string, []>("query_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_23_pad_0 = const()[name = tensor<string, []>("query_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_23_dilations_0 = const()[name = tensor<string, []>("query_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_23_groups_0 = const()[name = tensor<string, []>("query_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286972864)))];
            tensor<fp16, [1024]> layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289070080)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_23_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor<string, []>("query_23_cast_fp16")];
            tensor<string, []> key_23_pad_type_0 = const()[name = tensor<string, []>("key_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_23_strides_0 = const()[name = tensor<string, []>("key_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_23_pad_0 = const()[name = tensor<string, []>("key_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_23_dilations_0 = const()[name = tensor<string, []>("key_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_23_groups_0 = const()[name = tensor<string, []>("key_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289072192)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor<string, []>("key_23_cast_fp16")];
            tensor<string, []> value_23_pad_type_0 = const()[name = tensor<string, []>("value_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_23_strides_0 = const()[name = tensor<string, []>("value_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_23_pad_0 = const()[name = tensor<string, []>("value_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_23_dilations_0 = const()[name = tensor<string, []>("value_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_23_groups_0 = const()[name = tensor<string, []>("value_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291169408)))];
            tensor<fp16, [1024]> layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293266624)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor<string, []>("value_23_cast_fp16")];
            tensor<int32, [4]> var_1536 = const()[name = tensor<string, []>("op_1536"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_23_cast_fp16 = reshape(shape = var_1536, x = query_23_cast_fp16)[name = tensor<string, []>("mh_q_23_cast_fp16")];
            tensor<fp16, []> var_1538_to_fp16 = const()[name = tensor<string, []>("op_1538_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1539_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1538_to_fp16)[name = tensor<string, []>("op_1539_cast_fp16")];
            tensor<int32, [4]> var_1542 = const()[name = tensor<string, []>("op_1542"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1543_cast_fp16 = reshape(shape = var_1542, x = key_23_cast_fp16)[name = tensor<string, []>("op_1543_cast_fp16")];
            tensor<bool, []> mh_w_23_transpose_x_0 = const()[name = tensor<string, []>("mh_w_23_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_23_transpose_y_0 = const()[name = tensor<string, []>("mh_w_23_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_1539_cast_fp16, y = var_1543_cast_fp16)[name = tensor<string, []>("mh_w_23_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1546_cast_fp16 = softmax(axis = var_1478, x = mh_w_23_cast_fp16)[name = tensor<string, []>("op_1546_cast_fp16")];
            tensor<int32, [4]> var_1547 = const()[name = tensor<string, []>("op_1547"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1548_cast_fp16 = reshape(shape = var_1547, x = value_23_cast_fp16)[name = tensor<string, []>("op_1548_cast_fp16")];
            tensor<bool, []> attn_23_transpose_x_0 = const()[name = tensor<string, []>("attn_23_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_23_transpose_y_0 = const()[name = tensor<string, []>("attn_23_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1548_cast_fp16, y = var_1546_cast_fp16)[name = tensor<string, []>("attn_23_cast_fp16")];
            tensor<int32, [4]> var_1551 = const()[name = tensor<string, []>("op_1551"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_89_cast_fp16 = reshape(shape = var_1551, x = attn_23_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")];
            tensor<string, []> obj_47_pad_type_0 = const()[name = tensor<string, []>("obj_47_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_47_strides_0 = const()[name = tensor<string, []>("obj_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_47_pad_0 = const()[name = tensor<string, []>("obj_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_47_dilations_0 = const()[name = tensor<string, []>("obj_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_47_groups_0 = const()[name = tensor<string, []>("obj_47_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293268736)))];
            tensor<fp16, [1024]> layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295365952)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_47_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_47_dilations_0, groups = obj_47_groups_0, pad = obj_47_pad_0, pad_type = obj_47_pad_type_0, strides = obj_47_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = tensor<string, []>("obj_47_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_47_cast_fp16)[name = tensor<string, []>("inputs_47_cast_fp16")];
            tensor<int32, [1]> out_47_axes_0 = const()[name = tensor<string, []>("out_47_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1569_to_fp16 = const()[name = tensor<string, []>("op_1569_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1569_to_fp16, x = inputs_47_cast_fp16)[name = tensor<string, []>("out_47_cast_fp16")];
            tensor<fp16, [1024]> input_91_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_91_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295368064)))];
            tensor<fp16, [1024]> input_91_beta_0_to_fp16 = const()[name = tensor<string, []>("input_91_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295370176)))];
            tensor<fp16, []> input_91_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_91_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
            tensor<string, []> input_93_pad_type_0 = const()[name = tensor<string, []>("input_93_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_93_strides_0 = const()[name = tensor<string, []>("input_93_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_93_pad_0 = const()[name = tensor<string, []>("input_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_93_dilations_0 = const()[name = tensor<string, []>("input_93_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_93_groups_0 = const()[name = tensor<string, []>("input_93_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_11_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295372288)))];
            tensor<fp16, [4096]> layers_11_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303760960)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")];
            tensor<string, []> input_95_mode_0 = const()[name = tensor<string, []>("input_95_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = tensor<string, []>("input_95_cast_fp16")];
            tensor<string, []> hidden_states_27_pad_type_0 = const()[name = tensor<string, []>("hidden_states_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_27_strides_0 = const()[name = tensor<string, []>("hidden_states_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_27_pad_0 = const()[name = tensor<string, []>("hidden_states_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_27_dilations_0 = const()[name = tensor<string, []>("hidden_states_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_27_groups_0 = const()[name = tensor<string, []>("hidden_states_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_11_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303769216)))];
            tensor<fp16, [1024]> layers_11_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312157888)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_27_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = tensor<string, []>("hidden_states_27_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_27_cast_fp16)[name = tensor<string, []>("inputs_49_cast_fp16")];
            tensor<int32, []> var_1598 = const()[name = tensor<string, []>("op_1598"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_49_axes_0 = const()[name = tensor<string, []>("out_49_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1620_to_fp16 = const()[name = tensor<string, []>("op_1620_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1620_to_fp16, x = inputs_49_cast_fp16)[name = tensor<string, []>("out_49_cast_fp16")];
            tensor<fp16, [1024]> obj_49_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_49_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312160000)))];
            tensor<fp16, [1024]> obj_49_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_49_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312162112)))];
            tensor<fp16, []> obj_49_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_49_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_49_cast_fp16 = batch_norm(beta = obj_49_beta_0_to_fp16, epsilon = obj_49_epsilon_0_to_fp16, gamma = obj_49_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = tensor<string, []>("obj_49_cast_fp16")];
            tensor<string, []> query_25_pad_type_0 = const()[name = tensor<string, []>("query_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_25_strides_0 = const()[name = tensor<string, []>("query_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_25_pad_0 = const()[name = tensor<string, []>("query_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_25_dilations_0 = const()[name = tensor<string, []>("query_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_25_groups_0 = const()[name = tensor<string, []>("query_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312164224)))];
            tensor<fp16, [1024]> layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(314261440)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_25_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = tensor<string, []>("query_25_cast_fp16")];
            tensor<string, []> key_25_pad_type_0 = const()[name = tensor<string, []>("key_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_25_strides_0 = const()[name = tensor<string, []>("key_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_25_pad_0 = const()[name = tensor<string, []>("key_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_25_dilations_0 = const()[name = tensor<string, []>("key_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_25_groups_0 = const()[name = tensor<string, []>("key_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(314263552)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_25_cast_fp16 = conv(dilations = key_25_dilations_0, groups = key_25_groups_0, pad = key_25_pad_0, pad_type = key_25_pad_type_0, strides = key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = tensor<string, []>("key_25_cast_fp16")];
            tensor<string, []> value_25_pad_type_0 = const()[name = tensor<string, []>("value_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_25_strides_0 = const()[name = tensor<string, []>("value_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_25_pad_0 = const()[name = tensor<string, []>("value_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_25_dilations_0 = const()[name = tensor<string, []>("value_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_25_groups_0 = const()[name = tensor<string, []>("value_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316360768)))];
            tensor<fp16, [1024]> layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318457984)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = value_25_dilations_0, groups = value_25_groups_0, pad = value_25_pad_0, pad_type = value_25_pad_type_0, strides = value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = tensor<string, []>("value_25_cast_fp16")];
            tensor<int32, [4]> var_1656 = const()[name = tensor<string, []>("op_1656"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_25_cast_fp16 = reshape(shape = var_1656, x = query_25_cast_fp16)[name = tensor<string, []>("mh_q_25_cast_fp16")];
            tensor<fp16, []> var_1658_to_fp16 = const()[name = tensor<string, []>("op_1658_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1659_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1658_to_fp16)[name = tensor<string, []>("op_1659_cast_fp16")];
            tensor<int32, [4]> var_1662 = const()[name = tensor<string, []>("op_1662"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1663_cast_fp16 = reshape(shape = var_1662, x = key_25_cast_fp16)[name = tensor<string, []>("op_1663_cast_fp16")];
            tensor<bool, []> mh_w_25_transpose_x_0 = const()[name = tensor<string, []>("mh_w_25_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_25_transpose_y_0 = const()[name = tensor<string, []>("mh_w_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1659_cast_fp16, y = var_1663_cast_fp16)[name = tensor<string, []>("mh_w_25_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1666_cast_fp16 = softmax(axis = var_1598, x = mh_w_25_cast_fp16)[name = tensor<string, []>("op_1666_cast_fp16")];
            tensor<int32, [4]> var_1667 = const()[name = tensor<string, []>("op_1667"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1668_cast_fp16 = reshape(shape = var_1667, x = value_25_cast_fp16)[name = tensor<string, []>("op_1668_cast_fp16")];
            tensor<bool, []> attn_25_transpose_x_0 = const()[name = tensor<string, []>("attn_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_25_transpose_y_0 = const()[name = tensor<string, []>("attn_25_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1668_cast_fp16, y = var_1666_cast_fp16)[name = tensor<string, []>("attn_25_cast_fp16")];
            tensor<int32, [4]> var_1671 = const()[name = tensor<string, []>("op_1671"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_97_cast_fp16 = reshape(shape = var_1671, x = attn_25_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<string, []> obj_51_pad_type_0 = const()[name = tensor<string, []>("obj_51_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_51_strides_0 = const()[name = tensor<string, []>("obj_51_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_51_pad_0 = const()[name = tensor<string, []>("obj_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_51_dilations_0 = const()[name = tensor<string, []>("obj_51_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_51_groups_0 = const()[name = tensor<string, []>("obj_51_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318460096)))];
            tensor<fp16, [1024]> layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320557312)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_51_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_51_dilations_0, groups = obj_51_groups_0, pad = obj_51_pad_0, pad_type = obj_51_pad_type_0, strides = obj_51_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16, x = input_97_cast_fp16)[name = tensor<string, []>("obj_51_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_51_cast_fp16)[name = tensor<string, []>("inputs_51_cast_fp16")];
            tensor<int32, [1]> out_51_axes_0 = const()[name = tensor<string, []>("out_51_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1689_to_fp16 = const()[name = tensor<string, []>("op_1689_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_1689_to_fp16, x = inputs_51_cast_fp16)[name = tensor<string, []>("out_51_cast_fp16")];
            tensor<fp16, [1024]> input_99_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_99_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320559424)))];
            tensor<fp16, [1024]> input_99_beta_0_to_fp16 = const()[name = tensor<string, []>("input_99_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320561536)))];
            tensor<fp16, []> input_99_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_99_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_99_cast_fp16 = batch_norm(beta = input_99_beta_0_to_fp16, epsilon = input_99_epsilon_0_to_fp16, gamma = input_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")];
            tensor<string, []> input_101_pad_type_0 = const()[name = tensor<string, []>("input_101_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_101_strides_0 = const()[name = tensor<string, []>("input_101_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_101_pad_0 = const()[name = tensor<string, []>("input_101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_101_dilations_0 = const()[name = tensor<string, []>("input_101_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_101_groups_0 = const()[name = tensor<string, []>("input_101_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_12_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320563648)))];
            tensor<fp16, [4096]> layers_12_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328952320)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_101_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = layers_12_fc1_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")];
            tensor<string, []> input_103_mode_0 = const()[name = tensor<string, []>("input_103_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = input_101_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
            tensor<string, []> hidden_states_29_pad_type_0 = const()[name = tensor<string, []>("hidden_states_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_29_strides_0 = const()[name = tensor<string, []>("hidden_states_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_29_pad_0 = const()[name = tensor<string, []>("hidden_states_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_29_dilations_0 = const()[name = tensor<string, []>("hidden_states_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_29_groups_0 = const()[name = tensor<string, []>("hidden_states_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_12_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328960576)))];
            tensor<fp16, [1024]> layers_12_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337349248)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_29_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_12_fc2_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("hidden_states_29_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = hidden_states_29_cast_fp16)[name = tensor<string, []>("inputs_53_cast_fp16")];
            tensor<int32, []> var_1718 = const()[name = tensor<string, []>("op_1718"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_53_axes_0 = const()[name = tensor<string, []>("out_53_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1740_to_fp16 = const()[name = tensor<string, []>("op_1740_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_1740_to_fp16, x = inputs_53_cast_fp16)[name = tensor<string, []>("out_53_cast_fp16")];
            tensor<fp16, [1024]> obj_53_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_53_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337351360)))];
            tensor<fp16, [1024]> obj_53_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_53_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337353472)))];
            tensor<fp16, []> obj_53_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_53_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_53_cast_fp16 = batch_norm(beta = obj_53_beta_0_to_fp16, epsilon = obj_53_epsilon_0_to_fp16, gamma = obj_53_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = tensor<string, []>("obj_53_cast_fp16")];
            tensor<string, []> query_27_pad_type_0 = const()[name = tensor<string, []>("query_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_27_strides_0 = const()[name = tensor<string, []>("query_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_27_pad_0 = const()[name = tensor<string, []>("query_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_27_dilations_0 = const()[name = tensor<string, []>("query_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_27_groups_0 = const()[name = tensor<string, []>("query_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337355584)))];
            tensor<fp16, [1024]> layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339452800)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_27_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = tensor<string, []>("query_27_cast_fp16")];
            tensor<string, []> key_27_pad_type_0 = const()[name = tensor<string, []>("key_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_27_strides_0 = const()[name = tensor<string, []>("key_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_27_pad_0 = const()[name = tensor<string, []>("key_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_27_dilations_0 = const()[name = tensor<string, []>("key_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_27_groups_0 = const()[name = tensor<string, []>("key_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339454912)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = tensor<string, []>("key_27_cast_fp16")];
            tensor<string, []> value_27_pad_type_0 = const()[name = tensor<string, []>("value_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_27_strides_0 = const()[name = tensor<string, []>("value_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_27_pad_0 = const()[name = tensor<string, []>("value_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_27_dilations_0 = const()[name = tensor<string, []>("value_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_27_groups_0 = const()[name = tensor<string, []>("value_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341552128)))];
            tensor<fp16, [1024]> layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343649344)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = tensor<string, []>("value_27_cast_fp16")];
            tensor<int32, [4]> var_1776 = const()[name = tensor<string, []>("op_1776"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_27_cast_fp16 = reshape(shape = var_1776, x = query_27_cast_fp16)[name = tensor<string, []>("mh_q_27_cast_fp16")];
            tensor<fp16, []> var_1778_to_fp16 = const()[name = tensor<string, []>("op_1778_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1779_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1778_to_fp16)[name = tensor<string, []>("op_1779_cast_fp16")];
            tensor<int32, [4]> var_1782 = const()[name = tensor<string, []>("op_1782"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1783_cast_fp16 = reshape(shape = var_1782, x = key_27_cast_fp16)[name = tensor<string, []>("op_1783_cast_fp16")];
            tensor<bool, []> mh_w_27_transpose_x_0 = const()[name = tensor<string, []>("mh_w_27_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_27_transpose_y_0 = const()[name = tensor<string, []>("mh_w_27_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_27_cast_fp16 = matmul(transpose_x = mh_w_27_transpose_x_0, transpose_y = mh_w_27_transpose_y_0, x = var_1779_cast_fp16, y = var_1783_cast_fp16)[name = tensor<string, []>("mh_w_27_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1786_cast_fp16 = softmax(axis = var_1718, x = mh_w_27_cast_fp16)[name = tensor<string, []>("op_1786_cast_fp16")];
            tensor<int32, [4]> var_1787 = const()[name = tensor<string, []>("op_1787"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1788_cast_fp16 = reshape(shape = var_1787, x = value_27_cast_fp16)[name = tensor<string, []>("op_1788_cast_fp16")];
            tensor<bool, []> attn_27_transpose_x_0 = const()[name = tensor<string, []>("attn_27_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_27_transpose_y_0 = const()[name = tensor<string, []>("attn_27_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1788_cast_fp16, y = var_1786_cast_fp16)[name = tensor<string, []>("attn_27_cast_fp16")];
            tensor<int32, [4]> var_1791 = const()[name = tensor<string, []>("op_1791"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_105_cast_fp16 = reshape(shape = var_1791, x = attn_27_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
            tensor<string, []> obj_55_pad_type_0 = const()[name = tensor<string, []>("obj_55_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_55_strides_0 = const()[name = tensor<string, []>("obj_55_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_55_pad_0 = const()[name = tensor<string, []>("obj_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_55_dilations_0 = const()[name = tensor<string, []>("obj_55_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_55_groups_0 = const()[name = tensor<string, []>("obj_55_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343651456)))];
            tensor<fp16, [1024]> layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345748672)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_55_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_55_dilations_0, groups = obj_55_groups_0, pad = obj_55_pad_0, pad_type = obj_55_pad_type_0, strides = obj_55_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("obj_55_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_55_cast_fp16)[name = tensor<string, []>("inputs_55_cast_fp16")];
            tensor<int32, [1]> out_55_axes_0 = const()[name = tensor<string, []>("out_55_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1809_to_fp16 = const()[name = tensor<string, []>("op_1809_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_1809_to_fp16, x = inputs_55_cast_fp16)[name = tensor<string, []>("out_55_cast_fp16")];
            tensor<fp16, [1024]> input_107_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_107_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345750784)))];
            tensor<fp16, [1024]> input_107_beta_0_to_fp16 = const()[name = tensor<string, []>("input_107_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345752896)))];
            tensor<fp16, []> input_107_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_107_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_107_cast_fp16 = batch_norm(beta = input_107_beta_0_to_fp16, epsilon = input_107_epsilon_0_to_fp16, gamma = input_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")];
            tensor<string, []> input_109_pad_type_0 = const()[name = tensor<string, []>("input_109_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_109_strides_0 = const()[name = tensor<string, []>("input_109_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_109_pad_0 = const()[name = tensor<string, []>("input_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_109_dilations_0 = const()[name = tensor<string, []>("input_109_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_109_groups_0 = const()[name = tensor<string, []>("input_109_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_13_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345755008)))];
            tensor<fp16, [4096]> layers_13_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354143680)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_109_cast_fp16 = conv(bias = layers_13_fc1_bias_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 = layers_13_fc1_weight_to_fp16, x = input_107_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<string, []> input_111_mode_0 = const()[name = tensor<string, []>("input_111_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_111_cast_fp16 = gelu(mode = input_111_mode_0, x = input_109_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")];
            tensor<string, []> hidden_states_31_pad_type_0 = const()[name = tensor<string, []>("hidden_states_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_31_strides_0 = const()[name = tensor<string, []>("hidden_states_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_31_pad_0 = const()[name = tensor<string, []>("hidden_states_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_31_dilations_0 = const()[name = tensor<string, []>("hidden_states_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_31_groups_0 = const()[name = tensor<string, []>("hidden_states_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_13_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354151936)))];
            tensor<fp16, [1024]> layers_13_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362540608)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_31_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_13_fc2_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("hidden_states_31_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = hidden_states_31_cast_fp16)[name = tensor<string, []>("inputs_57_cast_fp16")];
            tensor<int32, []> var_1838 = const()[name = tensor<string, []>("op_1838"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_57_axes_0 = const()[name = tensor<string, []>("out_57_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1860_to_fp16 = const()[name = tensor<string, []>("op_1860_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_1860_to_fp16, x = inputs_57_cast_fp16)[name = tensor<string, []>("out_57_cast_fp16")];
            tensor<fp16, [1024]> obj_57_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_57_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362542720)))];
            tensor<fp16, [1024]> obj_57_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_57_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362544832)))];
            tensor<fp16, []> obj_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = tensor<string, []>("obj_57_cast_fp16")];
            tensor<string, []> query_29_pad_type_0 = const()[name = tensor<string, []>("query_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_29_strides_0 = const()[name = tensor<string, []>("query_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_29_pad_0 = const()[name = tensor<string, []>("query_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_29_dilations_0 = const()[name = tensor<string, []>("query_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_29_groups_0 = const()[name = tensor<string, []>("query_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362546944)))];
            tensor<fp16, [1024]> layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(364644160)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_29_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor<string, []>("query_29_cast_fp16")];
            tensor<string, []> key_29_pad_type_0 = const()[name = tensor<string, []>("key_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_29_strides_0 = const()[name = tensor<string, []>("key_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_29_pad_0 = const()[name = tensor<string, []>("key_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_29_dilations_0 = const()[name = tensor<string, []>("key_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_29_groups_0 = const()[name = tensor<string, []>("key_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(364646272)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_29_cast_fp16 = conv(dilations = key_29_dilations_0, groups = key_29_groups_0, pad = key_29_pad_0, pad_type = key_29_pad_type_0, strides = key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor<string, []>("key_29_cast_fp16")];
            tensor<string, []> value_29_pad_type_0 = const()[name = tensor<string, []>("value_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_29_strides_0 = const()[name = tensor<string, []>("value_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_29_pad_0 = const()[name = tensor<string, []>("value_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_29_dilations_0 = const()[name = tensor<string, []>("value_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_29_groups_0 = const()[name = tensor<string, []>("value_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(366743488)))];
            tensor<fp16, [1024]> layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368840704)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = value_29_dilations_0, groups = value_29_groups_0, pad = value_29_pad_0, pad_type = value_29_pad_type_0, strides = value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor<string, []>("value_29_cast_fp16")];
            tensor<int32, [4]> var_1896 = const()[name = tensor<string, []>("op_1896"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_29_cast_fp16 = reshape(shape = var_1896, x = query_29_cast_fp16)[name = tensor<string, []>("mh_q_29_cast_fp16")];
            tensor<fp16, []> var_1898_to_fp16 = const()[name = tensor<string, []>("op_1898_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_1899_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1898_to_fp16)[name = tensor<string, []>("op_1899_cast_fp16")];
            tensor<int32, [4]> var_1902 = const()[name = tensor<string, []>("op_1902"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1903_cast_fp16 = reshape(shape = var_1902, x = key_29_cast_fp16)[name = tensor<string, []>("op_1903_cast_fp16")];
            tensor<bool, []> mh_w_29_transpose_x_0 = const()[name = tensor<string, []>("mh_w_29_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_29_transpose_y_0 = const()[name = tensor<string, []>("mh_w_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1899_cast_fp16, y = var_1903_cast_fp16)[name = tensor<string, []>("mh_w_29_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_1906_cast_fp16 = softmax(axis = var_1838, x = mh_w_29_cast_fp16)[name = tensor<string, []>("op_1906_cast_fp16")];
            tensor<int32, [4]> var_1907 = const()[name = tensor<string, []>("op_1907"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1908_cast_fp16 = reshape(shape = var_1907, x = value_29_cast_fp16)[name = tensor<string, []>("op_1908_cast_fp16")];
            tensor<bool, []> attn_29_transpose_x_0 = const()[name = tensor<string, []>("attn_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_29_transpose_y_0 = const()[name = tensor<string, []>("attn_29_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1908_cast_fp16, y = var_1906_cast_fp16)[name = tensor<string, []>("attn_29_cast_fp16")];
            tensor<int32, [4]> var_1911 = const()[name = tensor<string, []>("op_1911"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_113_cast_fp16 = reshape(shape = var_1911, x = attn_29_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")];
            tensor<string, []> obj_59_pad_type_0 = const()[name = tensor<string, []>("obj_59_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_59_strides_0 = const()[name = tensor<string, []>("obj_59_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_59_pad_0 = const()[name = tensor<string, []>("obj_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_59_dilations_0 = const()[name = tensor<string, []>("obj_59_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_59_groups_0 = const()[name = tensor<string, []>("obj_59_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368842816)))];
            tensor<fp16, [1024]> layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370940032)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_59_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_59_dilations_0, groups = obj_59_groups_0, pad = obj_59_pad_0, pad_type = obj_59_pad_type_0, strides = obj_59_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = tensor<string, []>("obj_59_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_59_cast_fp16)[name = tensor<string, []>("inputs_59_cast_fp16")];
            tensor<int32, [1]> out_59_axes_0 = const()[name = tensor<string, []>("out_59_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1929_to_fp16 = const()[name = tensor<string, []>("op_1929_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_1929_to_fp16, x = inputs_59_cast_fp16)[name = tensor<string, []>("out_59_cast_fp16")];
            tensor<fp16, [1024]> input_115_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_115_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370942144)))];
            tensor<fp16, [1024]> input_115_beta_0_to_fp16 = const()[name = tensor<string, []>("input_115_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370944256)))];
            tensor<fp16, []> input_115_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_115_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<string, []> input_117_pad_type_0 = const()[name = tensor<string, []>("input_117_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_117_strides_0 = const()[name = tensor<string, []>("input_117_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_117_pad_0 = const()[name = tensor<string, []>("input_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_117_dilations_0 = const()[name = tensor<string, []>("input_117_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_117_groups_0 = const()[name = tensor<string, []>("input_117_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_14_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370946368)))];
            tensor<fp16, [4096]> layers_14_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(379335040)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_117_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_14_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")];
            tensor<string, []> input_119_mode_0 = const()[name = tensor<string, []>("input_119_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")];
            tensor<string, []> hidden_states_33_pad_type_0 = const()[name = tensor<string, []>("hidden_states_33_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_33_strides_0 = const()[name = tensor<string, []>("hidden_states_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_33_pad_0 = const()[name = tensor<string, []>("hidden_states_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_33_dilations_0 = const()[name = tensor<string, []>("hidden_states_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_33_groups_0 = const()[name = tensor<string, []>("hidden_states_33_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_14_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(379343296)))];
            tensor<fp16, [1024]> layers_14_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387731968)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_33_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_14_fc2_weight_to_fp16, x = input_119_cast_fp16)[name = tensor<string, []>("hidden_states_33_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_33_cast_fp16)[name = tensor<string, []>("inputs_61_cast_fp16")];
            tensor<int32, []> var_1958 = const()[name = tensor<string, []>("op_1958"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_61_axes_0 = const()[name = tensor<string, []>("out_61_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_1980_to_fp16 = const()[name = tensor<string, []>("op_1980_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_1980_to_fp16, x = inputs_61_cast_fp16)[name = tensor<string, []>("out_61_cast_fp16")];
            tensor<fp16, [1024]> obj_61_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_61_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387734080)))];
            tensor<fp16, [1024]> obj_61_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_61_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387736192)))];
            tensor<fp16, []> obj_61_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_61_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = tensor<string, []>("obj_61_cast_fp16")];
            tensor<string, []> query_31_pad_type_0 = const()[name = tensor<string, []>("query_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_31_strides_0 = const()[name = tensor<string, []>("query_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_31_pad_0 = const()[name = tensor<string, []>("query_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_31_dilations_0 = const()[name = tensor<string, []>("query_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_31_groups_0 = const()[name = tensor<string, []>("query_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387738304)))];
            tensor<fp16, [1024]> layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389835520)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_31_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = tensor<string, []>("query_31_cast_fp16")];
            tensor<string, []> key_31_pad_type_0 = const()[name = tensor<string, []>("key_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_31_strides_0 = const()[name = tensor<string, []>("key_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_31_pad_0 = const()[name = tensor<string, []>("key_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_31_dilations_0 = const()[name = tensor<string, []>("key_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_31_groups_0 = const()[name = tensor<string, []>("key_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389837632)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = tensor<string, []>("key_31_cast_fp16")];
            tensor<string, []> value_31_pad_type_0 = const()[name = tensor<string, []>("value_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_31_strides_0 = const()[name = tensor<string, []>("value_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_31_pad_0 = const()[name = tensor<string, []>("value_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_31_dilations_0 = const()[name = tensor<string, []>("value_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_31_groups_0 = const()[name = tensor<string, []>("value_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(391934848)))];
            tensor<fp16, [1024]> layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394032064)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = tensor<string, []>("value_31_cast_fp16")];
            tensor<int32, [4]> var_2016 = const()[name = tensor<string, []>("op_2016"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_31_cast_fp16 = reshape(shape = var_2016, x = query_31_cast_fp16)[name = tensor<string, []>("mh_q_31_cast_fp16")];
            tensor<fp16, []> var_2018_to_fp16 = const()[name = tensor<string, []>("op_2018_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2019_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_2018_to_fp16)[name = tensor<string, []>("op_2019_cast_fp16")];
            tensor<int32, [4]> var_2022 = const()[name = tensor<string, []>("op_2022"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2023_cast_fp16 = reshape(shape = var_2022, x = key_31_cast_fp16)[name = tensor<string, []>("op_2023_cast_fp16")];
            tensor<bool, []> mh_w_31_transpose_x_0 = const()[name = tensor<string, []>("mh_w_31_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_31_transpose_y_0 = const()[name = tensor<string, []>("mh_w_31_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_2019_cast_fp16, y = var_2023_cast_fp16)[name = tensor<string, []>("mh_w_31_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2026_cast_fp16 = softmax(axis = var_1958, x = mh_w_31_cast_fp16)[name = tensor<string, []>("op_2026_cast_fp16")];
            tensor<int32, [4]> var_2027 = const()[name = tensor<string, []>("op_2027"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2028_cast_fp16 = reshape(shape = var_2027, x = value_31_cast_fp16)[name = tensor<string, []>("op_2028_cast_fp16")];
            tensor<bool, []> attn_31_transpose_x_0 = const()[name = tensor<string, []>("attn_31_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_31_transpose_y_0 = const()[name = tensor<string, []>("attn_31_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_2028_cast_fp16, y = var_2026_cast_fp16)[name = tensor<string, []>("attn_31_cast_fp16")];
            tensor<int32, [4]> var_2031 = const()[name = tensor<string, []>("op_2031"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_121_cast_fp16 = reshape(shape = var_2031, x = attn_31_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")];
            tensor<string, []> obj_63_pad_type_0 = const()[name = tensor<string, []>("obj_63_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_63_strides_0 = const()[name = tensor<string, []>("obj_63_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_63_pad_0 = const()[name = tensor<string, []>("obj_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_63_dilations_0 = const()[name = tensor<string, []>("obj_63_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_63_groups_0 = const()[name = tensor<string, []>("obj_63_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394034176)))];
            tensor<fp16, [1024]> layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396131392)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_63_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16, x = input_121_cast_fp16)[name = tensor<string, []>("obj_63_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_63_cast_fp16)[name = tensor<string, []>("inputs_63_cast_fp16")];
            tensor<int32, [1]> out_63_axes_0 = const()[name = tensor<string, []>("out_63_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2049_to_fp16 = const()[name = tensor<string, []>("op_2049_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2049_to_fp16, x = inputs_63_cast_fp16)[name = tensor<string, []>("out_63_cast_fp16")];
            tensor<fp16, [1024]> input_123_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_123_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396133504)))];
            tensor<fp16, [1024]> input_123_beta_0_to_fp16 = const()[name = tensor<string, []>("input_123_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396135616)))];
            tensor<fp16, []> input_123_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_123_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_123_cast_fp16 = batch_norm(beta = input_123_beta_0_to_fp16, epsilon = input_123_epsilon_0_to_fp16, gamma = input_123_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")];
            tensor<string, []> input_125_pad_type_0 = const()[name = tensor<string, []>("input_125_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_125_strides_0 = const()[name = tensor<string, []>("input_125_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_125_pad_0 = const()[name = tensor<string, []>("input_125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_125_dilations_0 = const()[name = tensor<string, []>("input_125_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_125_groups_0 = const()[name = tensor<string, []>("input_125_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_15_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396137728)))];
            tensor<fp16, [4096]> layers_15_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(404526400)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_125_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_125_dilations_0, groups = input_125_groups_0, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = input_125_strides_0, weight = layers_15_fc1_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
            tensor<string, []> input_127_mode_0 = const()[name = tensor<string, []>("input_127_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_127_cast_fp16 = gelu(mode = input_127_mode_0, x = input_125_cast_fp16)[name = tensor<string, []>("input_127_cast_fp16")];
            tensor<string, []> hidden_states_35_pad_type_0 = const()[name = tensor<string, []>("hidden_states_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_35_strides_0 = const()[name = tensor<string, []>("hidden_states_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_35_pad_0 = const()[name = tensor<string, []>("hidden_states_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_35_dilations_0 = const()[name = tensor<string, []>("hidden_states_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_35_groups_0 = const()[name = tensor<string, []>("hidden_states_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_15_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(404534656)))];
            tensor<fp16, [1024]> layers_15_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412923328)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_35_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_15_fc2_weight_to_fp16, x = input_127_cast_fp16)[name = tensor<string, []>("hidden_states_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = hidden_states_35_cast_fp16)[name = tensor<string, []>("inputs_65_cast_fp16")];
            tensor<int32, []> var_2078 = const()[name = tensor<string, []>("op_2078"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_65_axes_0 = const()[name = tensor<string, []>("out_65_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2100_to_fp16 = const()[name = tensor<string, []>("op_2100_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2100_to_fp16, x = inputs_65_cast_fp16)[name = tensor<string, []>("out_65_cast_fp16")];
            tensor<fp16, [1024]> obj_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_65_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412925440)))];
            tensor<fp16, [1024]> obj_65_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_65_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412927552)))];
            tensor<fp16, []> obj_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = tensor<string, []>("obj_65_cast_fp16")];
            tensor<string, []> query_33_pad_type_0 = const()[name = tensor<string, []>("query_33_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_33_strides_0 = const()[name = tensor<string, []>("query_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_33_pad_0 = const()[name = tensor<string, []>("query_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_33_dilations_0 = const()[name = tensor<string, []>("query_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_33_groups_0 = const()[name = tensor<string, []>("query_33_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412929664)))];
            tensor<fp16, [1024]> layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415026880)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_33_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor<string, []>("query_33_cast_fp16")];
            tensor<string, []> key_33_pad_type_0 = const()[name = tensor<string, []>("key_33_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_33_strides_0 = const()[name = tensor<string, []>("key_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_33_pad_0 = const()[name = tensor<string, []>("key_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_33_dilations_0 = const()[name = tensor<string, []>("key_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_33_groups_0 = const()[name = tensor<string, []>("key_33_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415028992)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_33_cast_fp16 = conv(dilations = key_33_dilations_0, groups = key_33_groups_0, pad = key_33_pad_0, pad_type = key_33_pad_type_0, strides = key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor<string, []>("key_33_cast_fp16")];
            tensor<string, []> value_33_pad_type_0 = const()[name = tensor<string, []>("value_33_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_33_strides_0 = const()[name = tensor<string, []>("value_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_33_pad_0 = const()[name = tensor<string, []>("value_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_33_dilations_0 = const()[name = tensor<string, []>("value_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_33_groups_0 = const()[name = tensor<string, []>("value_33_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417126208)))];
            tensor<fp16, [1024]> layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419223424)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = value_33_dilations_0, groups = value_33_groups_0, pad = value_33_pad_0, pad_type = value_33_pad_type_0, strides = value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor<string, []>("value_33_cast_fp16")];
            tensor<int32, [4]> var_2136 = const()[name = tensor<string, []>("op_2136"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_33_cast_fp16 = reshape(shape = var_2136, x = query_33_cast_fp16)[name = tensor<string, []>("mh_q_33_cast_fp16")];
            tensor<fp16, []> var_2138_to_fp16 = const()[name = tensor<string, []>("op_2138_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2139_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_2138_to_fp16)[name = tensor<string, []>("op_2139_cast_fp16")];
            tensor<int32, [4]> var_2142 = const()[name = tensor<string, []>("op_2142"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2143_cast_fp16 = reshape(shape = var_2142, x = key_33_cast_fp16)[name = tensor<string, []>("op_2143_cast_fp16")];
            tensor<bool, []> mh_w_33_transpose_x_0 = const()[name = tensor<string, []>("mh_w_33_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_33_transpose_y_0 = const()[name = tensor<string, []>("mh_w_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_2139_cast_fp16, y = var_2143_cast_fp16)[name = tensor<string, []>("mh_w_33_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2146_cast_fp16 = softmax(axis = var_2078, x = mh_w_33_cast_fp16)[name = tensor<string, []>("op_2146_cast_fp16")];
            tensor<int32, [4]> var_2147 = const()[name = tensor<string, []>("op_2147"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2148_cast_fp16 = reshape(shape = var_2147, x = value_33_cast_fp16)[name = tensor<string, []>("op_2148_cast_fp16")];
            tensor<bool, []> attn_33_transpose_x_0 = const()[name = tensor<string, []>("attn_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_33_transpose_y_0 = const()[name = tensor<string, []>("attn_33_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_2148_cast_fp16, y = var_2146_cast_fp16)[name = tensor<string, []>("attn_33_cast_fp16")];
            tensor<int32, [4]> var_2151 = const()[name = tensor<string, []>("op_2151"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_129_cast_fp16 = reshape(shape = var_2151, x = attn_33_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")];
            tensor<string, []> obj_67_pad_type_0 = const()[name = tensor<string, []>("obj_67_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_67_strides_0 = const()[name = tensor<string, []>("obj_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_67_pad_0 = const()[name = tensor<string, []>("obj_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_67_dilations_0 = const()[name = tensor<string, []>("obj_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_67_groups_0 = const()[name = tensor<string, []>("obj_67_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419225536)))];
            tensor<fp16, [1024]> layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421322752)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_67_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16, x = input_129_cast_fp16)[name = tensor<string, []>("obj_67_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = obj_67_cast_fp16)[name = tensor<string, []>("inputs_67_cast_fp16")];
            tensor<int32, [1]> out_67_axes_0 = const()[name = tensor<string, []>("out_67_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2169_to_fp16 = const()[name = tensor<string, []>("op_2169_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2169_to_fp16, x = inputs_67_cast_fp16)[name = tensor<string, []>("out_67_cast_fp16")];
            tensor<fp16, [1024]> input_131_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_131_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421324864)))];
            tensor<fp16, [1024]> input_131_beta_0_to_fp16 = const()[name = tensor<string, []>("input_131_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421326976)))];
            tensor<fp16, []> input_131_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_131_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_131_cast_fp16 = batch_norm(beta = input_131_beta_0_to_fp16, epsilon = input_131_epsilon_0_to_fp16, gamma = input_131_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")];
            tensor<string, []> input_133_pad_type_0 = const()[name = tensor<string, []>("input_133_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_133_strides_0 = const()[name = tensor<string, []>("input_133_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_133_pad_0 = const()[name = tensor<string, []>("input_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_133_dilations_0 = const()[name = tensor<string, []>("input_133_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_133_groups_0 = const()[name = tensor<string, []>("input_133_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_16_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421329088)))];
            tensor<fp16, [4096]> layers_16_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429717760)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_133_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = layers_16_fc1_weight_to_fp16, x = input_131_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")];
            tensor<string, []> input_135_mode_0 = const()[name = tensor<string, []>("input_135_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_135_cast_fp16 = gelu(mode = input_135_mode_0, x = input_133_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")];
            tensor<string, []> hidden_states_37_pad_type_0 = const()[name = tensor<string, []>("hidden_states_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_37_strides_0 = const()[name = tensor<string, []>("hidden_states_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_37_pad_0 = const()[name = tensor<string, []>("hidden_states_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_37_dilations_0 = const()[name = tensor<string, []>("hidden_states_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_37_groups_0 = const()[name = tensor<string, []>("hidden_states_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_16_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429726016)))];
            tensor<fp16, [1024]> layers_16_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438114688)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_37_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_16_fc2_weight_to_fp16, x = input_135_cast_fp16)[name = tensor<string, []>("hidden_states_37_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = hidden_states_37_cast_fp16)[name = tensor<string, []>("inputs_69_cast_fp16")];
            tensor<int32, []> var_2198 = const()[name = tensor<string, []>("op_2198"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_69_axes_0 = const()[name = tensor<string, []>("out_69_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2220_to_fp16 = const()[name = tensor<string, []>("op_2220_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2220_to_fp16, x = inputs_69_cast_fp16)[name = tensor<string, []>("out_69_cast_fp16")];
            tensor<fp16, [1024]> obj_69_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_69_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438116800)))];
            tensor<fp16, [1024]> obj_69_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_69_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438118912)))];
            tensor<fp16, []> obj_69_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_69_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = tensor<string, []>("obj_69_cast_fp16")];
            tensor<string, []> query_35_pad_type_0 = const()[name = tensor<string, []>("query_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_35_strides_0 = const()[name = tensor<string, []>("query_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_35_pad_0 = const()[name = tensor<string, []>("query_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_35_dilations_0 = const()[name = tensor<string, []>("query_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_35_groups_0 = const()[name = tensor<string, []>("query_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438121024)))];
            tensor<fp16, [1024]> layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440218240)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_35_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = tensor<string, []>("query_35_cast_fp16")];
            tensor<string, []> key_35_pad_type_0 = const()[name = tensor<string, []>("key_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_35_strides_0 = const()[name = tensor<string, []>("key_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_35_pad_0 = const()[name = tensor<string, []>("key_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_35_dilations_0 = const()[name = tensor<string, []>("key_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_35_groups_0 = const()[name = tensor<string, []>("key_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440220352)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = tensor<string, []>("key_35_cast_fp16")];
            tensor<string, []> value_35_pad_type_0 = const()[name = tensor<string, []>("value_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_35_strides_0 = const()[name = tensor<string, []>("value_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_35_pad_0 = const()[name = tensor<string, []>("value_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_35_dilations_0 = const()[name = tensor<string, []>("value_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_35_groups_0 = const()[name = tensor<string, []>("value_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(442317568)))];
            tensor<fp16, [1024]> layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444414784)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = tensor<string, []>("value_35_cast_fp16")];
            tensor<int32, [4]> var_2256 = const()[name = tensor<string, []>("op_2256"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_35_cast_fp16 = reshape(shape = var_2256, x = query_35_cast_fp16)[name = tensor<string, []>("mh_q_35_cast_fp16")];
            tensor<fp16, []> var_2258_to_fp16 = const()[name = tensor<string, []>("op_2258_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2259_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_2258_to_fp16)[name = tensor<string, []>("op_2259_cast_fp16")];
            tensor<int32, [4]> var_2262 = const()[name = tensor<string, []>("op_2262"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2263_cast_fp16 = reshape(shape = var_2262, x = key_35_cast_fp16)[name = tensor<string, []>("op_2263_cast_fp16")];
            tensor<bool, []> mh_w_35_transpose_x_0 = const()[name = tensor<string, []>("mh_w_35_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_35_transpose_y_0 = const()[name = tensor<string, []>("mh_w_35_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_35_cast_fp16 = matmul(transpose_x = mh_w_35_transpose_x_0, transpose_y = mh_w_35_transpose_y_0, x = var_2259_cast_fp16, y = var_2263_cast_fp16)[name = tensor<string, []>("mh_w_35_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2266_cast_fp16 = softmax(axis = var_2198, x = mh_w_35_cast_fp16)[name = tensor<string, []>("op_2266_cast_fp16")];
            tensor<int32, [4]> var_2267 = const()[name = tensor<string, []>("op_2267"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2268_cast_fp16 = reshape(shape = var_2267, x = value_35_cast_fp16)[name = tensor<string, []>("op_2268_cast_fp16")];
            tensor<bool, []> attn_35_transpose_x_0 = const()[name = tensor<string, []>("attn_35_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_35_transpose_y_0 = const()[name = tensor<string, []>("attn_35_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_2268_cast_fp16, y = var_2266_cast_fp16)[name = tensor<string, []>("attn_35_cast_fp16")];
            tensor<int32, [4]> var_2271 = const()[name = tensor<string, []>("op_2271"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_137_cast_fp16 = reshape(shape = var_2271, x = attn_35_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")];
            tensor<string, []> obj_71_pad_type_0 = const()[name = tensor<string, []>("obj_71_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_71_strides_0 = const()[name = tensor<string, []>("obj_71_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_71_pad_0 = const()[name = tensor<string, []>("obj_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_71_dilations_0 = const()[name = tensor<string, []>("obj_71_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_71_groups_0 = const()[name = tensor<string, []>("obj_71_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444416896)))];
            tensor<fp16, [1024]> layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446514112)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_71_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_71_dilations_0, groups = obj_71_groups_0, pad = obj_71_pad_0, pad_type = obj_71_pad_type_0, strides = obj_71_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16, x = input_137_cast_fp16)[name = tensor<string, []>("obj_71_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_71_cast_fp16)[name = tensor<string, []>("inputs_71_cast_fp16")];
            tensor<int32, [1]> out_71_axes_0 = const()[name = tensor<string, []>("out_71_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2289_to_fp16 = const()[name = tensor<string, []>("op_2289_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2289_to_fp16, x = inputs_71_cast_fp16)[name = tensor<string, []>("out_71_cast_fp16")];
            tensor<fp16, [1024]> input_139_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_139_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446516224)))];
            tensor<fp16, [1024]> input_139_beta_0_to_fp16 = const()[name = tensor<string, []>("input_139_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446518336)))];
            tensor<fp16, []> input_139_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_139_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_139_cast_fp16 = batch_norm(beta = input_139_beta_0_to_fp16, epsilon = input_139_epsilon_0_to_fp16, gamma = input_139_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")];
            tensor<string, []> input_141_pad_type_0 = const()[name = tensor<string, []>("input_141_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_141_strides_0 = const()[name = tensor<string, []>("input_141_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_141_pad_0 = const()[name = tensor<string, []>("input_141_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_141_dilations_0 = const()[name = tensor<string, []>("input_141_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_141_groups_0 = const()[name = tensor<string, []>("input_141_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_17_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446520448)))];
            tensor<fp16, [4096]> layers_17_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454909120)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_141_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_141_dilations_0, groups = input_141_groups_0, pad = input_141_pad_0, pad_type = input_141_pad_type_0, strides = input_141_strides_0, weight = layers_17_fc1_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")];
            tensor<string, []> input_143_mode_0 = const()[name = tensor<string, []>("input_143_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_143_cast_fp16 = gelu(mode = input_143_mode_0, x = input_141_cast_fp16)[name = tensor<string, []>("input_143_cast_fp16")];
            tensor<string, []> hidden_states_39_pad_type_0 = const()[name = tensor<string, []>("hidden_states_39_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_39_strides_0 = const()[name = tensor<string, []>("hidden_states_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_39_pad_0 = const()[name = tensor<string, []>("hidden_states_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_39_dilations_0 = const()[name = tensor<string, []>("hidden_states_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_39_groups_0 = const()[name = tensor<string, []>("hidden_states_39_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_17_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454917376)))];
            tensor<fp16, [1024]> layers_17_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463306048)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_39_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_17_fc2_weight_to_fp16, x = input_143_cast_fp16)[name = tensor<string, []>("hidden_states_39_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_39_cast_fp16)[name = tensor<string, []>("inputs_73_cast_fp16")];
            tensor<int32, []> var_2318 = const()[name = tensor<string, []>("op_2318"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_73_axes_0 = const()[name = tensor<string, []>("out_73_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2340_to_fp16 = const()[name = tensor<string, []>("op_2340_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2340_to_fp16, x = inputs_73_cast_fp16)[name = tensor<string, []>("out_73_cast_fp16")];
            tensor<fp16, [1024]> obj_73_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_73_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463308160)))];
            tensor<fp16, [1024]> obj_73_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_73_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463310272)))];
            tensor<fp16, []> obj_73_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_73_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_73_cast_fp16 = batch_norm(beta = obj_73_beta_0_to_fp16, epsilon = obj_73_epsilon_0_to_fp16, gamma = obj_73_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = tensor<string, []>("obj_73_cast_fp16")];
            tensor<string, []> query_37_pad_type_0 = const()[name = tensor<string, []>("query_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_37_strides_0 = const()[name = tensor<string, []>("query_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_37_pad_0 = const()[name = tensor<string, []>("query_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_37_dilations_0 = const()[name = tensor<string, []>("query_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_37_groups_0 = const()[name = tensor<string, []>("query_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463312384)))];
            tensor<fp16, [1024]> layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465409600)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_37_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = tensor<string, []>("query_37_cast_fp16")];
            tensor<string, []> key_37_pad_type_0 = const()[name = tensor<string, []>("key_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_37_strides_0 = const()[name = tensor<string, []>("key_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_37_pad_0 = const()[name = tensor<string, []>("key_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_37_dilations_0 = const()[name = tensor<string, []>("key_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_37_groups_0 = const()[name = tensor<string, []>("key_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465411712)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_37_cast_fp16 = conv(dilations = key_37_dilations_0, groups = key_37_groups_0, pad = key_37_pad_0, pad_type = key_37_pad_type_0, strides = key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = tensor<string, []>("key_37_cast_fp16")];
            tensor<string, []> value_37_pad_type_0 = const()[name = tensor<string, []>("value_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_37_strides_0 = const()[name = tensor<string, []>("value_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_37_pad_0 = const()[name = tensor<string, []>("value_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_37_dilations_0 = const()[name = tensor<string, []>("value_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_37_groups_0 = const()[name = tensor<string, []>("value_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467508928)))];
            tensor<fp16, [1024]> layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(469606144)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = value_37_dilations_0, groups = value_37_groups_0, pad = value_37_pad_0, pad_type = value_37_pad_type_0, strides = value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = tensor<string, []>("value_37_cast_fp16")];
            tensor<int32, [4]> var_2376 = const()[name = tensor<string, []>("op_2376"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_37_cast_fp16 = reshape(shape = var_2376, x = query_37_cast_fp16)[name = tensor<string, []>("mh_q_37_cast_fp16")];
            tensor<fp16, []> var_2378_to_fp16 = const()[name = tensor<string, []>("op_2378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2379_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2378_to_fp16)[name = tensor<string, []>("op_2379_cast_fp16")];
            tensor<int32, [4]> var_2382 = const()[name = tensor<string, []>("op_2382"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2383_cast_fp16 = reshape(shape = var_2382, x = key_37_cast_fp16)[name = tensor<string, []>("op_2383_cast_fp16")];
            tensor<bool, []> mh_w_37_transpose_x_0 = const()[name = tensor<string, []>("mh_w_37_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_37_transpose_y_0 = const()[name = tensor<string, []>("mh_w_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_2379_cast_fp16, y = var_2383_cast_fp16)[name = tensor<string, []>("mh_w_37_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2386_cast_fp16 = softmax(axis = var_2318, x = mh_w_37_cast_fp16)[name = tensor<string, []>("op_2386_cast_fp16")];
            tensor<int32, [4]> var_2387 = const()[name = tensor<string, []>("op_2387"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2388_cast_fp16 = reshape(shape = var_2387, x = value_37_cast_fp16)[name = tensor<string, []>("op_2388_cast_fp16")];
            tensor<bool, []> attn_37_transpose_x_0 = const()[name = tensor<string, []>("attn_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_37_transpose_y_0 = const()[name = tensor<string, []>("attn_37_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2388_cast_fp16, y = var_2386_cast_fp16)[name = tensor<string, []>("attn_37_cast_fp16")];
            tensor<int32, [4]> var_2391 = const()[name = tensor<string, []>("op_2391"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_145_cast_fp16 = reshape(shape = var_2391, x = attn_37_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<string, []> obj_75_pad_type_0 = const()[name = tensor<string, []>("obj_75_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_75_strides_0 = const()[name = tensor<string, []>("obj_75_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_75_pad_0 = const()[name = tensor<string, []>("obj_75_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_75_dilations_0 = const()[name = tensor<string, []>("obj_75_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_75_groups_0 = const()[name = tensor<string, []>("obj_75_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(469608256)))];
            tensor<fp16, [1024]> layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471705472)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_75_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16, x = input_145_cast_fp16)[name = tensor<string, []>("obj_75_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_75_cast_fp16)[name = tensor<string, []>("inputs_75_cast_fp16")];
            tensor<int32, [1]> out_75_axes_0 = const()[name = tensor<string, []>("out_75_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2409_to_fp16 = const()[name = tensor<string, []>("op_2409_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_2409_to_fp16, x = inputs_75_cast_fp16)[name = tensor<string, []>("out_75_cast_fp16")];
            tensor<fp16, [1024]> input_147_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_147_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471707584)))];
            tensor<fp16, [1024]> input_147_beta_0_to_fp16 = const()[name = tensor<string, []>("input_147_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471709696)))];
            tensor<fp16, []> input_147_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_147_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")];
            tensor<string, []> input_149_pad_type_0 = const()[name = tensor<string, []>("input_149_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_149_strides_0 = const()[name = tensor<string, []>("input_149_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_149_pad_0 = const()[name = tensor<string, []>("input_149_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_149_dilations_0 = const()[name = tensor<string, []>("input_149_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_149_groups_0 = const()[name = tensor<string, []>("input_149_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_18_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471711808)))];
            tensor<fp16, [4096]> layers_18_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480100480)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_149_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_149_dilations_0, groups = input_149_groups_0, pad = input_149_pad_0, pad_type = input_149_pad_type_0, strides = input_149_strides_0, weight = layers_18_fc1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")];
            tensor<string, []> input_151_mode_0 = const()[name = tensor<string, []>("input_151_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")];
            tensor<string, []> hidden_states_41_pad_type_0 = const()[name = tensor<string, []>("hidden_states_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_41_strides_0 = const()[name = tensor<string, []>("hidden_states_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_41_pad_0 = const()[name = tensor<string, []>("hidden_states_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_41_dilations_0 = const()[name = tensor<string, []>("hidden_states_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_41_groups_0 = const()[name = tensor<string, []>("hidden_states_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_18_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480108736)))];
            tensor<fp16, [1024]> layers_18_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488497408)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_41_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_18_fc2_weight_to_fp16, x = input_151_cast_fp16)[name = tensor<string, []>("hidden_states_41_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = hidden_states_41_cast_fp16)[name = tensor<string, []>("inputs_77_cast_fp16")];
            tensor<int32, []> var_2438 = const()[name = tensor<string, []>("op_2438"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_77_axes_0 = const()[name = tensor<string, []>("out_77_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2460_to_fp16 = const()[name = tensor<string, []>("op_2460_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_2460_to_fp16, x = inputs_77_cast_fp16)[name = tensor<string, []>("out_77_cast_fp16")];
            tensor<fp16, [1024]> obj_77_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_77_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488499520)))];
            tensor<fp16, [1024]> obj_77_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_77_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488501632)))];
            tensor<fp16, []> obj_77_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_77_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_77_cast_fp16 = batch_norm(beta = obj_77_beta_0_to_fp16, epsilon = obj_77_epsilon_0_to_fp16, gamma = obj_77_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = tensor<string, []>("obj_77_cast_fp16")];
            tensor<string, []> query_39_pad_type_0 = const()[name = tensor<string, []>("query_39_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_39_strides_0 = const()[name = tensor<string, []>("query_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_39_pad_0 = const()[name = tensor<string, []>("query_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_39_dilations_0 = const()[name = tensor<string, []>("query_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_39_groups_0 = const()[name = tensor<string, []>("query_39_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488503744)))];
            tensor<fp16, [1024]> layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490600960)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_39_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = tensor<string, []>("query_39_cast_fp16")];
            tensor<string, []> key_39_pad_type_0 = const()[name = tensor<string, []>("key_39_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_39_strides_0 = const()[name = tensor<string, []>("key_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_39_pad_0 = const()[name = tensor<string, []>("key_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_39_dilations_0 = const()[name = tensor<string, []>("key_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_39_groups_0 = const()[name = tensor<string, []>("key_39_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490603072)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = tensor<string, []>("key_39_cast_fp16")];
            tensor<string, []> value_39_pad_type_0 = const()[name = tensor<string, []>("value_39_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_39_strides_0 = const()[name = tensor<string, []>("value_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_39_pad_0 = const()[name = tensor<string, []>("value_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_39_dilations_0 = const()[name = tensor<string, []>("value_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_39_groups_0 = const()[name = tensor<string, []>("value_39_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492700288)))];
            tensor<fp16, [1024]> layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494797504)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = tensor<string, []>("value_39_cast_fp16")];
            tensor<int32, [4]> var_2496 = const()[name = tensor<string, []>("op_2496"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_39_cast_fp16 = reshape(shape = var_2496, x = query_39_cast_fp16)[name = tensor<string, []>("mh_q_39_cast_fp16")];
            tensor<fp16, []> var_2498_to_fp16 = const()[name = tensor<string, []>("op_2498_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2499_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2498_to_fp16)[name = tensor<string, []>("op_2499_cast_fp16")];
            tensor<int32, [4]> var_2502 = const()[name = tensor<string, []>("op_2502"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2503_cast_fp16 = reshape(shape = var_2502, x = key_39_cast_fp16)[name = tensor<string, []>("op_2503_cast_fp16")];
            tensor<bool, []> mh_w_39_transpose_x_0 = const()[name = tensor<string, []>("mh_w_39_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_39_transpose_y_0 = const()[name = tensor<string, []>("mh_w_39_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_39_cast_fp16 = matmul(transpose_x = mh_w_39_transpose_x_0, transpose_y = mh_w_39_transpose_y_0, x = var_2499_cast_fp16, y = var_2503_cast_fp16)[name = tensor<string, []>("mh_w_39_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2506_cast_fp16 = softmax(axis = var_2438, x = mh_w_39_cast_fp16)[name = tensor<string, []>("op_2506_cast_fp16")];
            tensor<int32, [4]> var_2507 = const()[name = tensor<string, []>("op_2507"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2508_cast_fp16 = reshape(shape = var_2507, x = value_39_cast_fp16)[name = tensor<string, []>("op_2508_cast_fp16")];
            tensor<bool, []> attn_39_transpose_x_0 = const()[name = tensor<string, []>("attn_39_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_39_transpose_y_0 = const()[name = tensor<string, []>("attn_39_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2508_cast_fp16, y = var_2506_cast_fp16)[name = tensor<string, []>("attn_39_cast_fp16")];
            tensor<int32, [4]> var_2511 = const()[name = tensor<string, []>("op_2511"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_153_cast_fp16 = reshape(shape = var_2511, x = attn_39_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")];
            tensor<string, []> obj_79_pad_type_0 = const()[name = tensor<string, []>("obj_79_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_79_strides_0 = const()[name = tensor<string, []>("obj_79_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_79_pad_0 = const()[name = tensor<string, []>("obj_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_79_dilations_0 = const()[name = tensor<string, []>("obj_79_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_79_groups_0 = const()[name = tensor<string, []>("obj_79_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494799616)))];
            tensor<fp16, [1024]> layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496896832)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_79_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_79_dilations_0, groups = obj_79_groups_0, pad = obj_79_pad_0, pad_type = obj_79_pad_type_0, strides = obj_79_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16, x = input_153_cast_fp16)[name = tensor<string, []>("obj_79_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = obj_79_cast_fp16)[name = tensor<string, []>("inputs_79_cast_fp16")];
            tensor<int32, [1]> out_79_axes_0 = const()[name = tensor<string, []>("out_79_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2529_to_fp16 = const()[name = tensor<string, []>("op_2529_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_2529_to_fp16, x = inputs_79_cast_fp16)[name = tensor<string, []>("out_79_cast_fp16")];
            tensor<fp16, [1024]> input_155_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_155_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496898944)))];
            tensor<fp16, [1024]> input_155_beta_0_to_fp16 = const()[name = tensor<string, []>("input_155_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496901056)))];
            tensor<fp16, []> input_155_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_155_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")];
            tensor<string, []> input_157_pad_type_0 = const()[name = tensor<string, []>("input_157_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_157_strides_0 = const()[name = tensor<string, []>("input_157_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_157_pad_0 = const()[name = tensor<string, []>("input_157_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_157_dilations_0 = const()[name = tensor<string, []>("input_157_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_157_groups_0 = const()[name = tensor<string, []>("input_157_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_19_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496903168)))];
            tensor<fp16, [4096]> layers_19_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505291840)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_157_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_19_fc1_weight_to_fp16, x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<string, []> input_159_mode_0 = const()[name = tensor<string, []>("input_159_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = tensor<string, []>("input_159_cast_fp16")];
            tensor<string, []> hidden_states_43_pad_type_0 = const()[name = tensor<string, []>("hidden_states_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_43_strides_0 = const()[name = tensor<string, []>("hidden_states_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_43_pad_0 = const()[name = tensor<string, []>("hidden_states_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_43_dilations_0 = const()[name = tensor<string, []>("hidden_states_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_43_groups_0 = const()[name = tensor<string, []>("hidden_states_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_19_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505300096)))];
            tensor<fp16, [1024]> layers_19_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513688768)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_43_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_19_fc2_weight_to_fp16, x = input_159_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = hidden_states_43_cast_fp16)[name = tensor<string, []>("inputs_81_cast_fp16")];
            tensor<int32, []> var_2558 = const()[name = tensor<string, []>("op_2558"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_81_axes_0 = const()[name = tensor<string, []>("out_81_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2580_to_fp16 = const()[name = tensor<string, []>("op_2580_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_2580_to_fp16, x = inputs_81_cast_fp16)[name = tensor<string, []>("out_81_cast_fp16")];
            tensor<fp16, [1024]> obj_81_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_81_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513690880)))];
            tensor<fp16, [1024]> obj_81_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_81_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513692992)))];
            tensor<fp16, []> obj_81_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_81_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_81_cast_fp16 = batch_norm(beta = obj_81_beta_0_to_fp16, epsilon = obj_81_epsilon_0_to_fp16, gamma = obj_81_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = tensor<string, []>("obj_81_cast_fp16")];
            tensor<string, []> query_41_pad_type_0 = const()[name = tensor<string, []>("query_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_41_strides_0 = const()[name = tensor<string, []>("query_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_41_pad_0 = const()[name = tensor<string, []>("query_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_41_dilations_0 = const()[name = tensor<string, []>("query_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_41_groups_0 = const()[name = tensor<string, []>("query_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513695104)))];
            tensor<fp16, [1024]> layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(515792320)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_41_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = tensor<string, []>("query_41_cast_fp16")];
            tensor<string, []> key_41_pad_type_0 = const()[name = tensor<string, []>("key_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_41_strides_0 = const()[name = tensor<string, []>("key_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_41_pad_0 = const()[name = tensor<string, []>("key_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_41_dilations_0 = const()[name = tensor<string, []>("key_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_41_groups_0 = const()[name = tensor<string, []>("key_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(515794432)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_41_cast_fp16 = conv(dilations = key_41_dilations_0, groups = key_41_groups_0, pad = key_41_pad_0, pad_type = key_41_pad_type_0, strides = key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = tensor<string, []>("key_41_cast_fp16")];
            tensor<string, []> value_41_pad_type_0 = const()[name = tensor<string, []>("value_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_41_strides_0 = const()[name = tensor<string, []>("value_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_41_pad_0 = const()[name = tensor<string, []>("value_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_41_dilations_0 = const()[name = tensor<string, []>("value_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_41_groups_0 = const()[name = tensor<string, []>("value_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517891648)))];
            tensor<fp16, [1024]> layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(519988864)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = value_41_dilations_0, groups = value_41_groups_0, pad = value_41_pad_0, pad_type = value_41_pad_type_0, strides = value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = tensor<string, []>("value_41_cast_fp16")];
            tensor<int32, [4]> var_2616 = const()[name = tensor<string, []>("op_2616"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_41_cast_fp16 = reshape(shape = var_2616, x = query_41_cast_fp16)[name = tensor<string, []>("mh_q_41_cast_fp16")];
            tensor<fp16, []> var_2618_to_fp16 = const()[name = tensor<string, []>("op_2618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2619_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2618_to_fp16)[name = tensor<string, []>("op_2619_cast_fp16")];
            tensor<int32, [4]> var_2622 = const()[name = tensor<string, []>("op_2622"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2623_cast_fp16 = reshape(shape = var_2622, x = key_41_cast_fp16)[name = tensor<string, []>("op_2623_cast_fp16")];
            tensor<bool, []> mh_w_41_transpose_x_0 = const()[name = tensor<string, []>("mh_w_41_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_41_transpose_y_0 = const()[name = tensor<string, []>("mh_w_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_2619_cast_fp16, y = var_2623_cast_fp16)[name = tensor<string, []>("mh_w_41_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2626_cast_fp16 = softmax(axis = var_2558, x = mh_w_41_cast_fp16)[name = tensor<string, []>("op_2626_cast_fp16")];
            tensor<int32, [4]> var_2627 = const()[name = tensor<string, []>("op_2627"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2628_cast_fp16 = reshape(shape = var_2627, x = value_41_cast_fp16)[name = tensor<string, []>("op_2628_cast_fp16")];
            tensor<bool, []> attn_41_transpose_x_0 = const()[name = tensor<string, []>("attn_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_41_transpose_y_0 = const()[name = tensor<string, []>("attn_41_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2628_cast_fp16, y = var_2626_cast_fp16)[name = tensor<string, []>("attn_41_cast_fp16")];
            tensor<int32, [4]> var_2631 = const()[name = tensor<string, []>("op_2631"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_161_cast_fp16 = reshape(shape = var_2631, x = attn_41_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")];
            tensor<string, []> obj_83_pad_type_0 = const()[name = tensor<string, []>("obj_83_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_83_strides_0 = const()[name = tensor<string, []>("obj_83_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_83_pad_0 = const()[name = tensor<string, []>("obj_83_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_83_dilations_0 = const()[name = tensor<string, []>("obj_83_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_83_groups_0 = const()[name = tensor<string, []>("obj_83_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(519990976)))];
            tensor<fp16, [1024]> layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522088192)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_83_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_83_dilations_0, groups = obj_83_groups_0, pad = obj_83_pad_0, pad_type = obj_83_pad_type_0, strides = obj_83_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16, x = input_161_cast_fp16)[name = tensor<string, []>("obj_83_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_83_cast_fp16)[name = tensor<string, []>("inputs_83_cast_fp16")];
            tensor<int32, [1]> out_83_axes_0 = const()[name = tensor<string, []>("out_83_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2649_to_fp16 = const()[name = tensor<string, []>("op_2649_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_2649_to_fp16, x = inputs_83_cast_fp16)[name = tensor<string, []>("out_83_cast_fp16")];
            tensor<fp16, [1024]> input_163_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_163_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522090304)))];
            tensor<fp16, [1024]> input_163_beta_0_to_fp16 = const()[name = tensor<string, []>("input_163_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522092416)))];
            tensor<fp16, []> input_163_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_163_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_163_cast_fp16 = batch_norm(beta = input_163_beta_0_to_fp16, epsilon = input_163_epsilon_0_to_fp16, gamma = input_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")];
            tensor<string, []> input_165_pad_type_0 = const()[name = tensor<string, []>("input_165_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_165_strides_0 = const()[name = tensor<string, []>("input_165_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_165_pad_0 = const()[name = tensor<string, []>("input_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_165_dilations_0 = const()[name = tensor<string, []>("input_165_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_165_groups_0 = const()[name = tensor<string, []>("input_165_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_20_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522094528)))];
            tensor<fp16, [4096]> layers_20_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(530483200)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_165_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_165_dilations_0, groups = input_165_groups_0, pad = input_165_pad_0, pad_type = input_165_pad_type_0, strides = input_165_strides_0, weight = layers_20_fc1_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")];
            tensor<string, []> input_167_mode_0 = const()[name = tensor<string, []>("input_167_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_167_cast_fp16 = gelu(mode = input_167_mode_0, x = input_165_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")];
            tensor<string, []> hidden_states_45_pad_type_0 = const()[name = tensor<string, []>("hidden_states_45_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_45_strides_0 = const()[name = tensor<string, []>("hidden_states_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_45_pad_0 = const()[name = tensor<string, []>("hidden_states_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_45_dilations_0 = const()[name = tensor<string, []>("hidden_states_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_45_groups_0 = const()[name = tensor<string, []>("hidden_states_45_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_20_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(530491456)))];
            tensor<fp16, [1024]> layers_20_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538880128)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_45_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_20_fc2_weight_to_fp16, x = input_167_cast_fp16)[name = tensor<string, []>("hidden_states_45_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_45_cast_fp16)[name = tensor<string, []>("inputs_85_cast_fp16")];
            tensor<int32, []> var_2678 = const()[name = tensor<string, []>("op_2678"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_85_axes_0 = const()[name = tensor<string, []>("out_85_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2700_to_fp16 = const()[name = tensor<string, []>("op_2700_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_2700_to_fp16, x = inputs_85_cast_fp16)[name = tensor<string, []>("out_85_cast_fp16")];
            tensor<fp16, [1024]> obj_85_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_85_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538882240)))];
            tensor<fp16, [1024]> obj_85_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_85_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538884352)))];
            tensor<fp16, []> obj_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = tensor<string, []>("obj_85_cast_fp16")];
            tensor<string, []> query_43_pad_type_0 = const()[name = tensor<string, []>("query_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_43_strides_0 = const()[name = tensor<string, []>("query_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_43_pad_0 = const()[name = tensor<string, []>("query_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_43_dilations_0 = const()[name = tensor<string, []>("query_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_43_groups_0 = const()[name = tensor<string, []>("query_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538886464)))];
            tensor<fp16, [1024]> layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540983680)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_43_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor<string, []>("query_43_cast_fp16")];
            tensor<string, []> key_43_pad_type_0 = const()[name = tensor<string, []>("key_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_43_strides_0 = const()[name = tensor<string, []>("key_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_43_pad_0 = const()[name = tensor<string, []>("key_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_43_dilations_0 = const()[name = tensor<string, []>("key_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_43_groups_0 = const()[name = tensor<string, []>("key_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540985792)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor<string, []>("key_43_cast_fp16")];
            tensor<string, []> value_43_pad_type_0 = const()[name = tensor<string, []>("value_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_43_strides_0 = const()[name = tensor<string, []>("value_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_43_pad_0 = const()[name = tensor<string, []>("value_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_43_dilations_0 = const()[name = tensor<string, []>("value_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_43_groups_0 = const()[name = tensor<string, []>("value_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543083008)))];
            tensor<fp16, [1024]> layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545180224)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor<string, []>("value_43_cast_fp16")];
            tensor<int32, [4]> var_2736 = const()[name = tensor<string, []>("op_2736"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_43_cast_fp16 = reshape(shape = var_2736, x = query_43_cast_fp16)[name = tensor<string, []>("mh_q_43_cast_fp16")];
            tensor<fp16, []> var_2738_to_fp16 = const()[name = tensor<string, []>("op_2738_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2739_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2738_to_fp16)[name = tensor<string, []>("op_2739_cast_fp16")];
            tensor<int32, [4]> var_2742 = const()[name = tensor<string, []>("op_2742"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2743_cast_fp16 = reshape(shape = var_2742, x = key_43_cast_fp16)[name = tensor<string, []>("op_2743_cast_fp16")];
            tensor<bool, []> mh_w_43_transpose_x_0 = const()[name = tensor<string, []>("mh_w_43_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_43_transpose_y_0 = const()[name = tensor<string, []>("mh_w_43_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_43_cast_fp16 = matmul(transpose_x = mh_w_43_transpose_x_0, transpose_y = mh_w_43_transpose_y_0, x = var_2739_cast_fp16, y = var_2743_cast_fp16)[name = tensor<string, []>("mh_w_43_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2746_cast_fp16 = softmax(axis = var_2678, x = mh_w_43_cast_fp16)[name = tensor<string, []>("op_2746_cast_fp16")];
            tensor<int32, [4]> var_2747 = const()[name = tensor<string, []>("op_2747"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2748_cast_fp16 = reshape(shape = var_2747, x = value_43_cast_fp16)[name = tensor<string, []>("op_2748_cast_fp16")];
            tensor<bool, []> attn_43_transpose_x_0 = const()[name = tensor<string, []>("attn_43_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_43_transpose_y_0 = const()[name = tensor<string, []>("attn_43_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2748_cast_fp16, y = var_2746_cast_fp16)[name = tensor<string, []>("attn_43_cast_fp16")];
            tensor<int32, [4]> var_2751 = const()[name = tensor<string, []>("op_2751"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_169_cast_fp16 = reshape(shape = var_2751, x = attn_43_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")];
            tensor<string, []> obj_87_pad_type_0 = const()[name = tensor<string, []>("obj_87_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_87_strides_0 = const()[name = tensor<string, []>("obj_87_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_87_pad_0 = const()[name = tensor<string, []>("obj_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_87_dilations_0 = const()[name = tensor<string, []>("obj_87_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_87_groups_0 = const()[name = tensor<string, []>("obj_87_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545182336)))];
            tensor<fp16, [1024]> layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547279552)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_87_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_87_dilations_0, groups = obj_87_groups_0, pad = obj_87_pad_0, pad_type = obj_87_pad_type_0, strides = obj_87_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16, x = input_169_cast_fp16)[name = tensor<string, []>("obj_87_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_87_cast_fp16)[name = tensor<string, []>("inputs_87_cast_fp16")];
            tensor<int32, [1]> out_87_axes_0 = const()[name = tensor<string, []>("out_87_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2769_to_fp16 = const()[name = tensor<string, []>("op_2769_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_2769_to_fp16, x = inputs_87_cast_fp16)[name = tensor<string, []>("out_87_cast_fp16")];
            tensor<fp16, [1024]> input_171_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_171_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547281664)))];
            tensor<fp16, [1024]> input_171_beta_0_to_fp16 = const()[name = tensor<string, []>("input_171_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547283776)))];
            tensor<fp16, []> input_171_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_171_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_171_cast_fp16 = batch_norm(beta = input_171_beta_0_to_fp16, epsilon = input_171_epsilon_0_to_fp16, gamma = input_171_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")];
            tensor<string, []> input_173_pad_type_0 = const()[name = tensor<string, []>("input_173_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_173_strides_0 = const()[name = tensor<string, []>("input_173_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_173_pad_0 = const()[name = tensor<string, []>("input_173_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_173_dilations_0 = const()[name = tensor<string, []>("input_173_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_173_groups_0 = const()[name = tensor<string, []>("input_173_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_21_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547285888)))];
            tensor<fp16, [4096]> layers_21_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(555674560)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_173_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_173_dilations_0, groups = input_173_groups_0, pad = input_173_pad_0, pad_type = input_173_pad_type_0, strides = input_173_strides_0, weight = layers_21_fc1_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")];
            tensor<string, []> input_175_mode_0 = const()[name = tensor<string, []>("input_175_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = input_173_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<string, []> hidden_states_47_pad_type_0 = const()[name = tensor<string, []>("hidden_states_47_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_47_strides_0 = const()[name = tensor<string, []>("hidden_states_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_47_pad_0 = const()[name = tensor<string, []>("hidden_states_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_47_dilations_0 = const()[name = tensor<string, []>("hidden_states_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_47_groups_0 = const()[name = tensor<string, []>("hidden_states_47_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_21_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(555682816)))];
            tensor<fp16, [1024]> layers_21_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564071488)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_47_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_21_fc2_weight_to_fp16, x = input_175_cast_fp16)[name = tensor<string, []>("hidden_states_47_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = hidden_states_47_cast_fp16)[name = tensor<string, []>("inputs_89_cast_fp16")];
            tensor<int32, []> var_2798 = const()[name = tensor<string, []>("op_2798"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_89_axes_0 = const()[name = tensor<string, []>("out_89_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2820_to_fp16 = const()[name = tensor<string, []>("op_2820_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_2820_to_fp16, x = inputs_89_cast_fp16)[name = tensor<string, []>("out_89_cast_fp16")];
            tensor<fp16, [1024]> obj_89_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_89_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564073600)))];
            tensor<fp16, [1024]> obj_89_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_89_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564075712)))];
            tensor<fp16, []> obj_89_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_89_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_89_cast_fp16 = batch_norm(beta = obj_89_beta_0_to_fp16, epsilon = obj_89_epsilon_0_to_fp16, gamma = obj_89_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = tensor<string, []>("obj_89_cast_fp16")];
            tensor<string, []> query_45_pad_type_0 = const()[name = tensor<string, []>("query_45_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_45_strides_0 = const()[name = tensor<string, []>("query_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_45_pad_0 = const()[name = tensor<string, []>("query_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_45_dilations_0 = const()[name = tensor<string, []>("query_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_45_groups_0 = const()[name = tensor<string, []>("query_45_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564077824)))];
            tensor<fp16, [1024]> layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566175040)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_45_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = tensor<string, []>("query_45_cast_fp16")];
            tensor<string, []> key_45_pad_type_0 = const()[name = tensor<string, []>("key_45_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_45_strides_0 = const()[name = tensor<string, []>("key_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_45_pad_0 = const()[name = tensor<string, []>("key_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_45_dilations_0 = const()[name = tensor<string, []>("key_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_45_groups_0 = const()[name = tensor<string, []>("key_45_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566177152)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_45_cast_fp16 = conv(dilations = key_45_dilations_0, groups = key_45_groups_0, pad = key_45_pad_0, pad_type = key_45_pad_type_0, strides = key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = tensor<string, []>("key_45_cast_fp16")];
            tensor<string, []> value_45_pad_type_0 = const()[name = tensor<string, []>("value_45_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_45_strides_0 = const()[name = tensor<string, []>("value_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_45_pad_0 = const()[name = tensor<string, []>("value_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_45_dilations_0 = const()[name = tensor<string, []>("value_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_45_groups_0 = const()[name = tensor<string, []>("value_45_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568274368)))];
            tensor<fp16, [1024]> layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570371584)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = value_45_dilations_0, groups = value_45_groups_0, pad = value_45_pad_0, pad_type = value_45_pad_type_0, strides = value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = tensor<string, []>("value_45_cast_fp16")];
            tensor<int32, [4]> var_2856 = const()[name = tensor<string, []>("op_2856"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_45_cast_fp16 = reshape(shape = var_2856, x = query_45_cast_fp16)[name = tensor<string, []>("mh_q_45_cast_fp16")];
            tensor<fp16, []> var_2858_to_fp16 = const()[name = tensor<string, []>("op_2858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2859_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2858_to_fp16)[name = tensor<string, []>("op_2859_cast_fp16")];
            tensor<int32, [4]> var_2862 = const()[name = tensor<string, []>("op_2862"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2863_cast_fp16 = reshape(shape = var_2862, x = key_45_cast_fp16)[name = tensor<string, []>("op_2863_cast_fp16")];
            tensor<bool, []> mh_w_45_transpose_x_0 = const()[name = tensor<string, []>("mh_w_45_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_45_transpose_y_0 = const()[name = tensor<string, []>("mh_w_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_2859_cast_fp16, y = var_2863_cast_fp16)[name = tensor<string, []>("mh_w_45_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2866_cast_fp16 = softmax(axis = var_2798, x = mh_w_45_cast_fp16)[name = tensor<string, []>("op_2866_cast_fp16")];
            tensor<int32, [4]> var_2867 = const()[name = tensor<string, []>("op_2867"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2868_cast_fp16 = reshape(shape = var_2867, x = value_45_cast_fp16)[name = tensor<string, []>("op_2868_cast_fp16")];
            tensor<bool, []> attn_45_transpose_x_0 = const()[name = tensor<string, []>("attn_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_45_transpose_y_0 = const()[name = tensor<string, []>("attn_45_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2868_cast_fp16, y = var_2866_cast_fp16)[name = tensor<string, []>("attn_45_cast_fp16")];
            tensor<int32, [4]> var_2871 = const()[name = tensor<string, []>("op_2871"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_177_cast_fp16 = reshape(shape = var_2871, x = attn_45_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")];
            tensor<string, []> obj_91_pad_type_0 = const()[name = tensor<string, []>("obj_91_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_91_strides_0 = const()[name = tensor<string, []>("obj_91_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_91_pad_0 = const()[name = tensor<string, []>("obj_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_91_dilations_0 = const()[name = tensor<string, []>("obj_91_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_91_groups_0 = const()[name = tensor<string, []>("obj_91_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570373696)))];
            tensor<fp16, [1024]> layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572470912)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_91_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16, x = input_177_cast_fp16)[name = tensor<string, []>("obj_91_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = obj_91_cast_fp16)[name = tensor<string, []>("inputs_91_cast_fp16")];
            tensor<int32, [1]> out_91_axes_0 = const()[name = tensor<string, []>("out_91_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2889_to_fp16 = const()[name = tensor<string, []>("op_2889_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_2889_to_fp16, x = inputs_91_cast_fp16)[name = tensor<string, []>("out_91_cast_fp16")];
            tensor<fp16, [1024]> input_179_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_179_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572473024)))];
            tensor<fp16, [1024]> input_179_beta_0_to_fp16 = const()[name = tensor<string, []>("input_179_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572475136)))];
            tensor<fp16, []> input_179_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_179_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_179_cast_fp16 = batch_norm(beta = input_179_beta_0_to_fp16, epsilon = input_179_epsilon_0_to_fp16, gamma = input_179_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")];
            tensor<string, []> input_181_pad_type_0 = const()[name = tensor<string, []>("input_181_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_181_strides_0 = const()[name = tensor<string, []>("input_181_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_181_pad_0 = const()[name = tensor<string, []>("input_181_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_181_dilations_0 = const()[name = tensor<string, []>("input_181_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_181_groups_0 = const()[name = tensor<string, []>("input_181_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_22_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572477248)))];
            tensor<fp16, [4096]> layers_22_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(580865920)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_181_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_181_dilations_0, groups = input_181_groups_0, pad = input_181_pad_0, pad_type = input_181_pad_type_0, strides = input_181_strides_0, weight = layers_22_fc1_weight_to_fp16, x = input_179_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<string, []> input_183_mode_0 = const()[name = tensor<string, []>("input_183_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_183_cast_fp16 = gelu(mode = input_183_mode_0, x = input_181_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")];
            tensor<string, []> hidden_states_49_pad_type_0 = const()[name = tensor<string, []>("hidden_states_49_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_49_strides_0 = const()[name = tensor<string, []>("hidden_states_49_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_49_pad_0 = const()[name = tensor<string, []>("hidden_states_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_49_dilations_0 = const()[name = tensor<string, []>("hidden_states_49_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_49_groups_0 = const()[name = tensor<string, []>("hidden_states_49_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_22_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(580874176)))];
            tensor<fp16, [1024]> layers_22_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589262848)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_49_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_22_fc2_weight_to_fp16, x = input_183_cast_fp16)[name = tensor<string, []>("hidden_states_49_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = hidden_states_49_cast_fp16)[name = tensor<string, []>("inputs_93_cast_fp16")];
            tensor<int32, []> var_2918 = const()[name = tensor<string, []>("op_2918"), val = tensor<int32, []>(3)];
            tensor<int32, [1]> out_93_axes_0 = const()[name = tensor<string, []>("out_93_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_2940_to_fp16 = const()[name = tensor<string, []>("op_2940_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_2940_to_fp16, x = inputs_93_cast_fp16)[name = tensor<string, []>("out_93_cast_fp16")];
            tensor<fp16, [1024]> obj_93_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_93_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589264960)))];
            tensor<fp16, [1024]> obj_93_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_93_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589267072)))];
            tensor<fp16, []> obj_93_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_93_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = tensor<string, []>("obj_93_cast_fp16")];
            tensor<string, []> query_pad_type_0 = const()[name = tensor<string, []>("query_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> query_strides_0 = const()[name = tensor<string, []>("query_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_pad_0 = const()[name = tensor<string, []>("query_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_dilations_0 = const()[name = tensor<string, []>("query_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> query_groups_0 = const()[name = tensor<string, []>("query_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589269184)))];
            tensor<fp16, [1024]> layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591366400)))];
            tensor<fp16, [1, 1024, 1, 1500]> query_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor<string, []>("query_cast_fp16")];
            tensor<string, []> key_pad_type_0 = const()[name = tensor<string, []>("key_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> key_strides_0 = const()[name = tensor<string, []>("key_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_pad_0 = const()[name = tensor<string, []>("key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_dilations_0 = const()[name = tensor<string, []>("key_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> key_groups_0 = const()[name = tensor<string, []>("key_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591368512)))];
            tensor<fp16, [1, 1024, 1, 1500]> key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor<string, []>("key_cast_fp16")];
            tensor<string, []> value_pad_type_0 = const()[name = tensor<string, []>("value_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> value_strides_0 = const()[name = tensor<string, []>("value_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_pad_0 = const()[name = tensor<string, []>("value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_dilations_0 = const()[name = tensor<string, []>("value_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> value_groups_0 = const()[name = tensor<string, []>("value_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(593465728)))];
            tensor<fp16, [1024]> layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595562944)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor<string, []>("value_cast_fp16")];
            tensor<int32, [4]> var_2976 = const()[name = tensor<string, []>("op_2976"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> mh_q_cast_fp16 = reshape(shape = var_2976, x = query_cast_fp16)[name = tensor<string, []>("mh_q_cast_fp16")];
            tensor<fp16, []> var_2978_to_fp16 = const()[name = tensor<string, []>("op_2978_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1500]> var_2979_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_2978_to_fp16)[name = tensor<string, []>("op_2979_cast_fp16")];
            tensor<int32, [4]> var_2982 = const()[name = tensor<string, []>("op_2982"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2983_cast_fp16 = reshape(shape = var_2982, x = key_cast_fp16)[name = tensor<string, []>("op_2983_cast_fp16")];
            tensor<bool, []> mh_w_transpose_x_0 = const()[name = tensor<string, []>("mh_w_transpose_x_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> mh_w_transpose_y_0 = const()[name = tensor<string, []>("mh_w_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 1500]> mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_2979_cast_fp16, y = var_2983_cast_fp16)[name = tensor<string, []>("mh_w_cast_fp16")];
            tensor<fp16, [1, 16, 1500, 1500]> var_2986_cast_fp16 = softmax(axis = var_2918, x = mh_w_cast_fp16)[name = tensor<string, []>("op_2986_cast_fp16")];
            tensor<int32, [4]> var_2987 = const()[name = tensor<string, []>("op_2987"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2988_cast_fp16 = reshape(shape = var_2987, x = value_cast_fp16)[name = tensor<string, []>("op_2988_cast_fp16")];
            tensor<bool, []> attn_transpose_x_0 = const()[name = tensor<string, []>("attn_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_transpose_y_0 = const()[name = tensor<string, []>("attn_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 16, 64, 1500]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_2988_cast_fp16, y = var_2986_cast_fp16)[name = tensor<string, []>("attn_cast_fp16")];
            tensor<int32, [4]> var_2991 = const()[name = tensor<string, []>("op_2991"), val = tensor<int32, [4]>([1, 1024, 1, 1500])];
            tensor<fp16, [1, 1024, 1, 1500]> input_185_cast_fp16 = reshape(shape = var_2991, x = attn_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")];
            tensor<string, []> obj_pad_type_0 = const()[name = tensor<string, []>("obj_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> obj_strides_0 = const()[name = tensor<string, []>("obj_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_pad_0 = const()[name = tensor<string, []>("obj_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_dilations_0 = const()[name = tensor<string, []>("obj_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> obj_groups_0 = const()[name = tensor<string, []>("obj_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595565056)))];
            tensor<fp16, [1024]> layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597662272)))];
            tensor<fp16, [1, 1024, 1, 1500]> obj_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16, x = input_185_cast_fp16)[name = tensor<string, []>("obj_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_cast_fp16)[name = tensor<string, []>("inputs_95_cast_fp16")];
            tensor<int32, [1]> out_95_axes_0 = const()[name = tensor<string, []>("out_95_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_3009_to_fp16 = const()[name = tensor<string, []>("op_3009_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_3009_to_fp16, x = inputs_95_cast_fp16)[name = tensor<string, []>("out_95_cast_fp16")];
            tensor<fp16, [1024]> input_187_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_187_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597664384)))];
            tensor<fp16, [1024]> input_187_beta_0_to_fp16 = const()[name = tensor<string, []>("input_187_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597666496)))];
            tensor<fp16, []> input_187_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_187_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> input_187_cast_fp16 = batch_norm(beta = input_187_beta_0_to_fp16, epsilon = input_187_epsilon_0_to_fp16, gamma = input_187_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")];
            tensor<string, []> input_189_pad_type_0 = const()[name = tensor<string, []>("input_189_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_189_strides_0 = const()[name = tensor<string, []>("input_189_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_189_pad_0 = const()[name = tensor<string, []>("input_189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_189_dilations_0 = const()[name = tensor<string, []>("input_189_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_189_groups_0 = const()[name = tensor<string, []>("input_189_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_23_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597668608)))];
            tensor<fp16, [4096]> layers_23_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606057280)))];
            tensor<fp16, [1, 4096, 1, 1500]> input_189_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = layers_23_fc1_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")];
            tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 1500]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_189_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
            tensor<string, []> hidden_states_pad_type_0 = const()[name = tensor<string, []>("hidden_states_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> hidden_states_strides_0 = const()[name = tensor<string, []>("hidden_states_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_pad_0 = const()[name = tensor<string, []>("hidden_states_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_dilations_0 = const()[name = tensor<string, []>("hidden_states_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> hidden_states_groups_0 = const()[name = tensor<string, []>("hidden_states_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_23_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606065536)))];
            tensor<fp16, [1024]> layers_23_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614454208)))];
            tensor<fp16, [1, 1024, 1, 1500]> hidden_states_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_23_fc2_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("hidden_states_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1500]> inputs_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")];
            tensor<int32, [1]> out_axes_0 = const()[name = tensor<string, []>("out_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, []> var_3047_to_fp16 = const()[name = tensor<string, []>("op_3047_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_3047_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("out_cast_fp16")];
            tensor<fp16, [1024]> encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = tensor<string, []>("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614456320)))];
            tensor<fp16, [1024]> encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = tensor<string, []>("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614458432)))];
            tensor<fp16, []> encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = tensor<string, []>("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1500]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = tensor<string, []>("encoder_output_embeds_type_fp32_cast_fp16")];
        } -> (encoder_output_embeds);
}