pytorch/benchmark

Pull Requests Commits

davidberard98 committed 3 years ago

5ae9de18

Add switches for turning on/off inductor cudagraphs & DDP static graphs

davidberard98 committed 3 years ago

72486594

unlandable changes

davidberard98 committed 3 years ago

f6085294

Update DDP result parser for gathering partial results (#1251)

davidberard98 committed 3 years ago

5b046066

Use torch._dynamo instead of torchdynamo (#1252)

davidberard98 committed 3 years ago

78bc9f0b

ddp experiments: run all measurements on the same allocation (#1245)

davidberard98 committed 3 years ago

d2ebc396

Fix aicluster ci (#1250)

xuzhao9 committed 3 years ago

121a5584

Add 1.13.0rc test config, cu116 as a start (#1234)

weiwangmeta committed 3 years ago

6089f4ea

Allow equal_nan in correctness check

bertmaher committed 3 years ago

341ef123

correctness check train

jspark1105 committed 3 years ago

65e4f449

Upgrade to checkout@v3 (#1240)

xuzhao9 committed 3 years ago

e403c97e

Required changes after cuBLAS workspace allocation change (#1226)

eqy committed 3 years ago

c5879304

Fix name of the platform (#1241)

xuzhao9 committed 3 years ago

5a89750b

Add PYTORCH_TEST_SKIP_CUDA_MEM_LEAK_CHECK (#1237)

sangongs committed 3 years ago

f5f795bb

ddp+dynamo: assorted fixes (#1223)

davidberard98 committed 3 years ago

f320cda1

Implemented staged tests for train

xuzhao9 committed 3 years ago

9ca9f58e

Enable CI on AWS T4 Metal instance (#1236)

xuzhao9 committed 3 years ago

d6a3e064

functorch userbenchmark (#1235)

zou3519 committed 3 years ago

147390dc

disable torchinductor.config.triton.use_bmm with tritonmm=triton option

jspark1105 committed 3 years ago

5814dbcc

Migrate instruction count benchmark to userbenchmark (#1195)

xuzhao9 committed 3 years ago

1ea5f44c

Fix broken ci from gpt2_large (#1230)

xuzhao9 committed 3 years ago

20686776

Add flag for dynamo+ddp optimizations (#1221)

davidberard98 committed 3 years ago

c8133956

Pin pycocotools version to make the patch work (#1224)

xuzhao9 committed 3 years ago

d0bf8a65

change training precision default to tf32

jspark1105 committed 3 years ago

628ee78c

Another fix for the bisection job (#1227)

xuzhao9 committed 3 years ago

45f0678e

Update ddp+dynamo benchmark arguments (#1222)

davidberard98 committed 3 years ago

48ca7613

Load instance setup script before running benchmark (#1217)

xuzhao9 committed 3 years ago

b96d07a9

Add torchinductor triton config mm

Mengchi Zhang committed 3 years ago

c3813be9

add infra and models for ddp experiments (#1190)

wconstab committed 3 years ago

37e714b0

Support customized code evaluation in torchdynamo options

xuzhao9 committed 3 years ago

b18d4f6f

Older