pytorch/benchmark

Pull Requests Commits

[optim][ub] Exclude yolov3 from NAdam with PT2 on no_foreach and differentiable

janeyx99 committed 2 years ago

98bb866a

Skip installing pytorch nightly. (#1607)

xuzhao9 committed 2 years ago

f2ca46f0

Fix docker push (#1601)

xuzhao9 committed 2 years ago

22244cd3

Add regression detection steps for torch-nightly (#1603)

xuzhao9 committed 2 years ago

d1eba6ae

Fix a major bug (#1606)

janeyx99 committed 2 years ago

3c8b3304

Use dynamo's compile_times() API to get compilation time metrics

bertmaher committed 2 years ago

73871abc

force a dynamo cache clean with gc.collect to fix OOMs (#1600)

janeyx99 committed 2 years ago

76a4d8bd

Add push to the docker workflow (#1596)

xuzhao9 committed 2 years ago

f9f893ea

Skip most pt2 forloop impls, add option to ignore skips (#1594)

janeyx99 committed 2 years ago

b26d77a6

Guard target dir check for non-Nones (#1593)

janeyx99 committed 2 years ago

cf03e87c

Add workflow and scripts to build torchbench nightly docker (#1588)

xuzhao9 committed 2 years ago

e1eae7a9

Support customizing batch size in train test (#1505)

xuzhao9 committed 2 years ago

bb00321d

Check the nightly versions are correct after installation (#1591)

xuzhao9 committed 2 years ago

4fefc47b

Fix the return value of select_device() (#1586)

xuzhao9 committed 2 years ago

0a6151db

Add more exceptions, improve -s and -m interaction (#1583)

janeyx99 committed 2 years ago

1b139f3f

Give the option to write the metrics json elsewhere (#1580)

janeyx99 committed 2 years ago

c9b46856

Run userbenchmarks in subprocess in CI (#1584)

janeyx99 committed 2 years ago

fe9e92d0

remove asserts when selecting cuda device (#1585)

janeyx99 committed 2 years ago

da10d3b7

Fix the broken CI by disabling warmup by default (#1582)

xuzhao9 committed 2 years ago

9baf1358

Remove duplication, add typing, allow more functionality in userbenchmark utils (#1579)

janeyx99 committed 2 years ago

db849f40

Support userbenchmarks to inject and replace the `model.invoke()` function (#1574)

xuzhao9 committed 2 years ago

4a1df233

Add torch.no_grad() to inference tests (#1577)

xuzhao9 committed 2 years ago

8620df66

Add torchaudio nightly install step for build compatibility (#1576)

spzala committed 2 years ago

5b882382

Add compilation time support (#1547)

xuzhao9 committed 2 years ago

118a28ba

Fix a subset of issues mentioned in #1567 (#1573)

janeyx99 committed 2 years ago

9b7cfc5b

Update the main branch of pytorch (#1570)

xuzhao9 committed 2 years ago

50720510

Fix the batch size for maml_omniglot and speech_transformer (#1569)

xuzhao9 committed 2 years ago

7f33d625

Enable amp support (BFloat16) in CPU (#1516)

yanbing-j committed 2 years ago

147a0494

Add option `none` to disable all metrics (#1565)

FindHao committed 2 years ago

a8e532b1

Add new GPU arch supports for FMA count (#1568)

FindHao committed 2 years ago

3b538611

Older