pytorch/benchmark

Pull Requests Commits

comment out transformer_auto_wrapper_policy while we investigate why it degrades perf

erichan1 committed 3 years ago

98344682

add transformer sharding policy to fsdp in t5 distributed benchmark

erichan1 committed 3 years ago

229505ba

Merge branch 'main' of https://github.com/pytorch/benchmark into erichan1/t5-fsdp

erichan1 committed 3 years ago

38953ab4

Remove the use of niter in the model code (#1171)

xuzhao9 committed 3 years ago

72bee5cd

Move workflows to CUDA 11.6 (#1162)

xuzhao9 committed 3 years ago

3b91b5a6

Fix a device inconsistency in dlrm (#1170)

desertfire committed 3 years ago

24b95f2f

Check type of example inputs from model.get_module() (#1167)

nmacchioni committed 3 years ago

e02e7ff0

Deploy the distributed CI (#1161)

xuzhao9 committed 3 years ago

7b4f223b

Checkout file from S3 instead of git lfs (#1158)

xuzhao9 committed 3 years ago

44a0ab0c

Install the most recent succesfull nightly builds in Circle CI (#1164)

xuzhao9 committed 3 years ago

8004157b

Rewrite the torchdynamo optimizer to use decorator (#1165)

xuzhao9 committed 3 years ago

60257569

Adding missing self.loader (#1148)

anijain2305 committed 3 years ago

b145cbf1

Fix pytorch_unet model (#1146)

xuzhao9 committed 3 years ago

111ff640

erichan1 committed 3 years ago

325fb44f

move distributed before optimizer

erichan1 committed 3 years ago

804fe7c3

Merge branch 'main' of https://github.com/pytorch/benchmark into erichan1/t5-fsdp

erichan1 committed 3 years ago

781bb7f2

fix unet script problem (#1092)

zzpmiracle committed 3 years ago

05a84d99

Revert changes to distributed/trainer.py from #1135 (#1144)

davidberard98 committed 3 years ago

76e464cc

Setup CUDA environment when running bisection and abtest (#1141)

xuzhao9 committed 3 years ago

c678454d

initial attempt at adding fsdp to T5

erichan1 committed 3 years ago

2c5849cb

distributed submit script fixes for model_args (#1135)

davidberard98 committed 3 years ago

2fc9a573

Add T5 e2e_benchmark translation task with support for DDP (#1122)

erichan1 committed 3 years ago

e89c2b32

Add run scripts for the distributed CI on AWS cluster (#1127)

xuzhao9 committed 3 years ago

758c9ee6

Add CUDA version comparison workflow (#1132)

xuzhao9 committed 3 years ago

a166e123

Minor fixes for distributed metrics (#1130)

wconstab committed 3 years ago

9ee26050

Fix backward-compatiblity issue when mps was not available in pytorch (#1129)

xuzhao9 committed 3 years ago

c7fd92c0

accept 2 outputs from parse_torchdynamo_args (#1128)

davidberard98 committed 3 years ago

1ec85e20

Also turn on JIT profiling mode (#1126)

davidberard98 committed 3 years ago

ba67ad70

Setup nvfuser automatic nightly CI (#1124)

xuzhao9 committed 3 years ago

031b028f

arg name fix (#1125)

zzpmiracle committed 3 years ago

e0392965

Older