pytorch/benchmark

Pull Requests Commits

Enable sam_fast

huydhn committed 1 year ago

d158caf3

Undo hf_Roberta_base change

huydhn committed 1 year ago

887905bd

Let's keep hf_T5_generate as it is

huydhn committed 1 year ago

69311f4e

Fix hf_T5_generate

huydhn committed 1 year ago

28fa2a24

hf_Roberta_base failed accuracy on A10G

huydhn committed 1 year ago

93961440

Update torchbench pin on PyTorch CI

huydhn committed 1 year ago

09fbfe32

Log guard latency (#145132)

anijain2305 committed 1 year ago

be908b3d

Increase tolerance for amp only poolformer_m36 (#145375)

Ivan Kobzarev committed 1 year ago

4b95e3ab

update compile time benchmarks to dump compile times to stdout and csv (#145447)

xmfan committed 1 year ago

c19f8243

Fix mobilenetv2 inductor freezing fail_accuracy (#145296)

Ivan Kobzarev committed 1 year ago

412a0f3d

Add `--profile-details` and `--export-perfdoctor` option (#2565)

xuzhao9 committed 1 year ago

b52426e5

PEP585 update - torch/_dynamo (#145105)

aorenste committed 1 year ago

7a38a39d

PEP585 update - benchmarks tools torchgen (#145101)

aorenste committed 1 year ago

cc21a4e6

Update/Fix userbenchmark-a100.yml (#2571)

atalman committed 1 year ago

1bdb1f31

dynamo: Don't crash with internal error if getattr on a tensor fails (#144817)

c00w committed 1 year ago

7a2a2f67

Use the manifold cache backend as the default (#144773)

AishwaryaSivaraman committed 1 year ago

68c54098

Implement increment and add_to_set for CompileEventLogger

jamesjwu committed 1 year ago

d0835775

Migrate from Tuple -> tuple in torch/_dynamo (#144261)

bobrenjc93 committed 1 year ago

9b7cac9d

easy: sort dictionary keys for inductor config when publishing (#143307)

c00w committed 1 year ago

e7f85d27

Update upload-artifact action to v4 (#2564)

kit1980 committed 1 year ago

af94f0e6

fix ruff rule E226: add missing whitespace around operator in f-strings (#144415)

XuehaiPan committed 1 year ago

cfe2709e

Remove unnecessary copy of gradients in util (#144329)

Skylion007 committed 1 year ago

d15a3cc5

Introduce cache hot loading APIs (a.k.a. "Mega-cache") (#143341)

oulgen committed 1 year ago

31cc1a49

Migrate from Tuple -> tuple in benchmarks (#144259)

bobrenjc93 committed 1 year ago

1e7ed466

LazyVT utils to get original value/source and is_hashable (#144160)

anijain2305 committed 1 year ago

9028c93f

Move symnode helpers to utils (#144158)

anijain2305 committed 1 year ago

916f107b

Introduce CompileEventLogger, replace usages of metrics_context and chromium_event with it

jamesjwu committed 1 year ago

2229ce5d

Propagate callable parameter types using ParamSpec (#142306) (#143797)

kaspell committed 1 year ago

f574229c

Remove DICT_SUBCLASS_GUARD_MANAGER and use dict.keys (#143722)

anijain2305 committed 1 year ago

c06c4896

A few fixes/updates to record_compilation_metrics (#143332)

masnesral committed 1 year ago

379bea5a

Older