Open TedThemistokleous opened 1 week ago
TODO:
Attention: Patch coverage is 40.00000%
with 12 lines
in your changes missing coverage. Please review.
Project coverage is 91.99%. Comparing base (
9dcea5c
) to head (7f62a33
). Report is 3 commits behind head on develop.
Files with missing lines | Patch % | Lines |
---|---|---|
src/onnx/parse_matmul.cpp | 40.00% | 12 Missing :warning: |
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Test | Batch | Rate new f912e6 |
Rate old ab5da3 |
Diff | Compare | |
---|---|---|---|---|---|---|
torchvision-resnet50 | 64 | 3,250.83 | 3,250.20 | 0.02% | :white_check_mark: | |
torchvision-resnet50_fp16 | 64 | 6,989.30 | 6,981.72 | 0.11% | :white_check_mark: | |
torchvision-densenet121 | 32 | 2,432.92 | 2,434.35 | -0.06% | :white_check_mark: | |
torchvision-densenet121_fp16 | 32 | 4,051.61 | 4,104.80 | -1.30% | :white_check_mark: | |
torchvision-inceptionv3 | 32 | 1,637.62 | 1,637.98 | -0.02% | :white_check_mark: | |
torchvision-inceptionv3_fp16 | 32 | 2,743.04 | 2,743.88 | -0.03% | :white_check_mark: | |
cadene-inceptionv4 | 16 | 778.79 | 778.58 | 0.03% | :white_check_mark: | |
cadene-resnext64x4 | 16 | 808.70 | 808.70 | 0.00% | :white_check_mark: | |
slim-mobilenet | 64 | 7,458.65 | 7,457.95 | 0.01% | :white_check_mark: | |
slim-nasnetalarge | 64 | 208.49 | 208.52 | -0.02% | :white_check_mark: | |
slim-resnet50v2 | 64 | 3,436.87 | 3,435.85 | 0.03% | :white_check_mark: | |
bert-mrpc-onnx | 8 | 1,149.67 | 1,153.18 | -0.30% | :white_check_mark: | |
bert-mrpc-tf | 1 | 307.53 | 305.52 | 0.66% | :white_check_mark: | |
pytorch-examples-wlang-gru | 1 | 519.82 | 384.89 | 35.05% | :high_brightness: | |
pytorch-examples-wlang-lstm | 1 | 396.01 | 386.28 | 2.52% | :white_check_mark: | |
torchvision-resnet50_1 | 1 | 773.75 | 768.57 | 0.67% | :white_check_mark: | |
cadene-dpn92_1 | 1 | 399.79 | 437.45 | -8.61% | :red_circle: | |
cadene-resnext101_1 | 1 | 381.87 | 383.50 | -0.42% | :white_check_mark: | |
onnx-taau-downsample | 1 | 344.15 | 344.79 | -0.18% | :white_check_mark: | |
dlrm-criteoterabyte | 1 | 35.04 | 35.03 | 0.03% | :white_check_mark: | |
dlrm-criteoterabyte_fp16 | 1 | 58.08 | 58.06 | 0.04% | :white_check_mark: | |
agentmodel | 1 | 7,873.77 | 7,924.68 | -0.64% | :white_check_mark: | |
unet_fp16 | 2 | 58.06 | 57.94 | 0.22% | :white_check_mark: | |
resnet50v1_fp16 | 1 | 959.34 | 929.40 | 3.22% | :high_brightness: | |
resnet50v1_int8 | 1 | 965.21 | 960.30 | 0.51% | :white_check_mark: | |
bert_base_cased_fp16 | 64 | 1,153.71 | 1,153.43 | 0.02% | :white_check_mark: | |
bert_large_uncased_fp16 | 32 | 356.06 | 355.83 | 0.07% | :white_check_mark: | |
bert_large_fp16 | 1 | 211.93 | 210.44 | 0.71% | :white_check_mark: | |
distilgpt2_fp16 | 16 | 2,158.70 | 2,159.20 | -0.02% | :white_check_mark: | |
yolov5s | 1 | 539.46 | 539.05 | 0.08% | :white_check_mark: | |
tinyllama | 1 | 43.38 | 43.40 | -0.04% | :white_check_mark: | |
vicuna-fastchat | 1 | 176.02 | 170.95 | 2.96% | :white_check_mark: | |
whisper-tiny-encoder | 1 | 417.79 | 417.95 | -0.04% | :white_check_mark: | |
whisper-tiny-decoder | 1 | 422.13 | 423.97 | -0.43% | :white_check_mark: |
This build is not recommended to merge :red_circle:
:red_circle:bert_large_uncased_fp16: FAILED: MIGraphX is not within tolerance - check verbose output
Changes to MatMul parser to handle the Microsoft Contrib operator MatMulintegarToFloat
Since we have the scale and zero points in our operands we can just perform a multiplied after int8 biases are added and then insert a regular dot on the scaled input values which should give the same output as the input data types.
Able to leverage the existing set of tests for matmul