pytorch/benchmark

Pull Requests Commits

Adding support for hpu device (#2595)

arathi-hlab committed 337 days ago

12a7578b

Record how many parameters we're parsing within dynamo (#148508)

masnesral committed 342 days ago

87d954eb

Propagate callable parameter types using ParamSpec (#142306) (#151014)

generatedunixname499836121 committed 343 days ago

2269ff7f

handle tensor subclass with non-classmethod `__torch_function__` (#151061)

StrongerXi committed 344 days ago

8448045c

Avoid using fp32 in sdp math path when benchmark performance. (#150996)

generatedunixname499836121 committed 344 days ago

97b78e1b

don't return logits for benchmark script (#151075)

shunting314 committed 344 days ago

edf32331

Restore functionality of t4 metal and c5.24xlarge workflows (#2610)

atalman committed 349 days ago

2a7fdcaa

reconstruct functions decorated in the compiled region properly (#150645)

williamwen42 committed 351 days ago

6eb17f1e

Add pgo remote get/put timings to dynamo_compile (#150322)

masnesral committed 352 days ago

98b06f00

Fix mis-calculated memory compression ratio (#150695)

desertfire committed 352 days ago

de0cb2f6

Fix `dict.items()` return type (#150112)

generatedunixname499836121 committed 355 days ago

4dbceeab

Make CompileEventLogger more defensive w.r.t to AOTAutogradCache and FXGraphCache (#150423)

jamesjwu committed 355 days ago

df6622bb

Update how peak memory is measured (#150534)

desertfire committed 356 days ago

942979ef

Always trace into tensor subclass `__torch_function__` (#149792)

StrongerXi committed 357 days ago

e5c91641

add dynamo disable reasons to codebase (#150440)

williamwen42 committed 357 days ago

5186143e

add APIs to determine a class is a namedtuple or PyStructSequence (#113257)

generatedunixname499836121 committed 357 days ago

8fae8cca

Revert D71711852

Dark Knight committed 358 days ago

40a841b6

always set deterministic for xpu accuracy test (#149028)

generatedunixname499836121 committed 359 days ago

77276109

Fix _Waitcounter decorator and dd backward pass wait counter (#150235)

ppanchalia committed 360 days ago

cf97e29e

Fix `is_compile_supported()` when `device_type` contains device index (#147837)

generatedunixname499836121 committed 362 days ago

d71863d0

Add --output-iter-metrics flag to cpu userbenchmark scripts (#2600)

murste01 committed 364 days ago

49c3f18f

Fix handling of setattr with some tensor attributes (#149791)

StrongerXi committed 1 year ago

a2b6092c

use torch.compile ca API for benchmarks (#149647)

xmfan committed 1 year ago

7481a407

Unify `cuBLASLt` workspaces with `cuBLAS` workspaces (#145130)

generatedunixname499836121 committed 1 year ago

90e73750

Add python version to dynamo_compile table (#149419)

masnesral committed 1 year ago

10a7be34

Switch off inference mode during compilation (#149321)

anijain2305 committed 1 year ago

2c5bc4ad

fix dynamic_shapes spec for moco (#148772) (#2601)

pianpwk committed 1 year ago

d1b2abbf

fix two accuracy regression (#149172)

shunting314 committed 1 year ago

50e2f744

NotImplementedError: Model's DEFAULT_TRAIN_BSIZE is not implemented. (#2563)

ostrowskimarcin committed 1 year ago

4dc08945

Set compile_id in the CachingAutotuner during compilation so we have it for dynamo_timed logging (#148693)

masnesral committed 1 year ago

ee4aa8ba

Newer Older