vision-bnn-benchmarks-hf
/
checkpoints
/shallow998_reproduceMaxim000_nas_8b
/statistics_hardware_checkpoint
Top: | |
epoch | |
extras | |
state_dict | |
arch | |
------------------------------------- | |
arch: ai85nascifarnet | |
------------------------------------- | |
extras: {'best epoch': 123456789, 'best_top1': 64.78999853134155, 'clipping_method': 'MAX_BIT_SHIFT', 'current_top1': 64.78999853134155} | |
------------------------------------- | |
state_dict: | |
conv1_1 | |
output_shift: [-0.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 64 , [64] | |
# of unique elements: 63 | |
min, max, mean: -1568.0 , 1831.0 , 38.640625 | |
weight | |
total # of elements, shape: 1728 , [64, 3, 3, 3] | |
# of unique elements: 155 | |
min, max, mean: -105.0 , 109.0 , -0.122106485 | |
conv1_2 | |
output_shift: [-1.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 32 , [32] | |
# of unique elements: 32 | |
min, max, mean: -15481.0 , 7800.0 , 3626.5625 | |
weight | |
total # of elements, shape: 2048 , [32, 64, 1, 1] | |
# of unique elements: 200 | |
min, max, mean: -128.0 , 127.0 , -5.270508 | |
conv1_3 | |
output_shift: [-1.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 64 , [64] | |
# of unique elements: 63 | |
min, max, mean: -7690.0 , 16383.0 , 3146.8281 | |
weight | |
total # of elements, shape: 18432 , [64, 32, 3, 3] | |
# of unique elements: 184 | |
min, max, mean: -116.0 , 103.0 , -1.2205946 | |
conv2_1 | |
output_shift: [-3.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 32 , [32] | |
# of unique elements: 10 | |
min, max, mean: -16384.0 , 16383.0 , 1596.0625 | |
weight | |
total # of elements, shape: 18432 , [32, 64, 3, 3] | |
# of unique elements: 232 | |
min, max, mean: -128.0 , 127.0 , 0.12641059 | |
conv2_2 | |
output_shift: [-0.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 64 , [64] | |
# of unique elements: 64 | |
min, max, mean: -3721.0 , 11407.0 , 964.0 | |
weight | |
total # of elements, shape: 2048 , [64, 32, 1, 1] | |
# of unique elements: 209 | |
min, max, mean: -128.0 , 127.0 , -2.3925781 | |
conv3_1 | |
output_shift: [-3.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 128 , [128] | |
# of unique elements: 47 | |
min, max, mean: -16384.0 , 16383.0 , 3604.875 | |
weight | |
total # of elements, shape: 73728 , [128, 64, 3, 3] | |
# of unique elements: 250 | |
min, max, mean: -128.0 , 127.0 , -0.31487358 | |
conv3_2 | |
output_shift: [-1.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 128 , [128] | |
# of unique elements: 122 | |
min, max, mean: -10467.0 , 16383.0 , 4923.797 | |
weight | |
total # of elements, shape: 16384 , [128, 128, 1, 1] | |
# of unique elements: 234 | |
min, max, mean: -128.0 , 127.0 , -3.060669 | |
conv4_1 | |
output_shift: [-3.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 64 , [64] | |
# of unique elements: 16 | |
min, max, mean: -16384.0 , 16383.0 , -2837.8125 | |
weight | |
total # of elements, shape: 73728 , [64, 128, 3, 3] | |
# of unique elements: 228 | |
min, max, mean: -125.0 , 127.0 , 0.3362766 | |
conv4_2 | |
output_shift: [-2.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 128 , [128] | |
# of unique elements: 128 | |
min, max, mean: -15940.0 , 16383.0 , 3565.0312 | |
weight | |
total # of elements, shape: 73728 , [128, 64, 3, 3] | |
# of unique elements: 244 | |
min, max, mean: -128.0 , 127.0 , -0.21457249 | |
conv5_1 | |
output_shift: [-1.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 128 , [128] | |
# of unique elements: 128 | |
min, max, mean: -9980.0 , 16383.0 , 3269.8203 | |
weight | |
total # of elements, shape: 16384 , [128, 128, 1, 1] | |
# of unique elements: 212 | |
min, max, mean: -108.0 , 127.0 , 0.20965576 | |
fc | |
output_shift: [1.] | |
adjust_output_shift: [0.] | |
quantize_activation: [1.] | |
shift_quantile: [0.995] | |
weight bits: [8.] | |
bias_bits: [8.] | |
bias | |
total # of elements, shape: 100 , [100] | |
# of unique elements: 99 | |
min, max, mean: -1737.0 , 1383.0 , -2.73 | |
weight | |
total # of elements, shape: 51200 , [100, 512] | |
# of unique elements: 212 | |
min, max, mean: -128.0 , 115.0 , -10.220196 | |