pytorch/benchmark

Pull Requests Commits

Merge branch 'main' into wwei9/fix-attention

Wei Wei committed 3 years ago

b5eac28a

Remove numba deps (#811)

xuzhao9 committed 3 years ago

aabd309f

NVFuser microbenchmarks and filtering tool (#793)

davidberard98 committed 3 years ago

67d08eac

Revert D34929680: Multisect successfully blamed D34929680 for test failures (#74381)

Dark Knight committed 3 years ago

5b38fc02

Disable correctness ci because it is flaky (#807)

xuzhao9 committed 3 years ago

cad410ef

Replacing fp16 and int8 mode with enum type (#74338)

Mengchi Zhang committed 3 years ago

b5d4646c

Enable fp16 half on timm models (#802)

xuzhao9 committed 3 years ago

c6df90c1

jit scripting has trouble handling finfo

Wei Wei committed 3 years ago

abb0f633

Checkout the main branch before finding the correct commit on date. (#804)

xuzhao9 committed 3 years ago

bff77519

add some explanation to help people understand

Wei Wei committed 3 years ago

b1a8d8c0

use torch sys float16 minimum; fix an importing issue

Wei Wei committed 3 years ago

215e2324

type change to float

Wei Wei committed 3 years ago

a705a4fe

Clean up the leaked files in the root directory (#792)

xuzhao9 committed 3 years ago

d39385de

Merge branch 'main' into wwei9/fix-attention

Wei Wei committed 3 years ago

aaec19ec

Fix bisection workflow broken by vision and ffmpeg upgrade (#797)

xuzhao9 committed 3 years ago

2a8528f9

Implement unicode_csv_reader because it was removed in upstream. (#795)

xuzhao9 committed 3 years ago

59b1758f

attention issue fix for fp16 inference

Wei Wei committed 3 years ago

f1224741

Enable CPU test on speech_transformer (for accuracy test) (#791)

xuzhao9 committed 3 years ago

06beb921

Add CPU implementation to pytorch_CycleGAN_and_pix2pix (#790)

xuzhao9 committed 3 years ago

0f641a08

Remove model version lock and test on Python 3.9 (#787)

xuzhao9 committed 3 years ago

7a6ab934

Fixes Background_Matting get_module interface (#785)

xuzhao9 committed 3 years ago

dc94612d

Enable fp16 by default for hf models (#782)

xuzhao9 committed 3 years ago

536969ca

Use fuser contextmanager (#784)

xuzhao9 committed 3 years ago

93a4a7fb

Remove "NotImplemented" flags if the test is limited by hardware capacity (#781)

xuzhao9 committed 3 years ago

f2c83718

Enable fx2trt on Huggingface models (#778)

xuzhao9 committed 3 years ago

ee89c449

Add correctness CI. (#779)

xuzhao9 committed 3 years ago

99086b33

Add fp16 amp mode support for all models (#776)

xuzhao9 committed 3 years ago

cc4d824d

Add the hf_bert E2E benchmark (#771)

xuzhao9 committed 3 years ago

3d672dde

Factor the core lowering component into a pass function (#777)

kflu committed 3 years ago

c714a6e2

Fix a env check bug for ondemand ci (#770)

xuzhao9 committed 3 years ago

4bf35cf7

Older