strisunshinewentingwang/pytorch

Pull Requests Commits

Vectorize the softmax calculation when not along the last dim (#59195)

leslie-fang-intel committed 3 years ago

68d690ff

Make PyObject_FastGetAttrString accept const char* (#59758)

ezyang committed 3 years ago

d60d81b5

Fix expecttest accept on Python 3.8 and later (#59709)

ezyang committed 3 years ago

700add07

Alias for `digamma` as `psi` to `special` namespace (#59143)

krshrimali committed 3 years ago

cf38b20c

Update on "[DDP] Support for multiple backwards"

rohan-varma committed 3 years ago

b638ca73

Update base for Update on "[DDP] Support for multiple backwards"

rohan-varma committed 3 years ago

97743d99

Improve numerical stability of GroupNorm (#54921)

xiaomengy committed 3 years ago

ff15d93b

Update on "[DDP] Support for multiple backwards"

rohan-varma committed 3 years ago

7845e104

Update base for Update on "[DDP] Support for multiple backwards"

rohan-varma committed 3 years ago

a920ac5f

MemoryOverlap: Avoid has_storage calls (#59013)

peterbell10 committed 3 years ago

095cd6a0

[CUDA graphs] Make stream semantics of backward calls consistent with other cuda ops (ci-all edition) (#57833)

mcarilli committed 3 years ago

be038d89

Update on "[DDP] Support for multiple backwards"

rohan-varma committed 3 years ago

5ac9c5d4

Update base for Update on "[DDP] Support for multiple backwards"

rohan-varma committed 3 years ago

77de42e7

Revert D28994140: [pytorch][PR] Implemented torch.cov

Mike Ruberry committed 3 years ago

92513038

Refactor SavedVariable (#59836)

Varal7 committed 3 years ago

0ceea7fa

pre compute regex and match simple signature autograd codegen 15s -> 12s (#59852)

albanD committed 3 years ago

d03ff1a1

refactor yaml loader import, no runtime change (#59850)

albanD committed 3 years ago

30a18fe3

Use C dumper if possible aten codegen 23s -> 13s (#59849)

albanD committed 3 years ago

c60d1ac9

avoid error string formatting aten codegen 28s -> 23s (#59848)

albanD committed 3 years ago

504ec301

Avoid unnecessary re-computation autograd codegen 21s -> 15s (#59847)

albanD committed 3 years ago

7143a6a1

Simplify parametrizations.SpectralNorm and improve its initialization (#59564)

lezcano committed 3 years ago

1f6e3933

Fix bad change in a CUDACachingAllocator loop (#59903)

r-barnes committed 3 years ago

10a3a3d3

Automated submodule update: FBGEMM (#59874)

facebook-github-bot committed 3 years ago

e49f0f4f

Revert D28981326: torch/lib/c10d: Use torch_check instead of throwing runtime_error

rohan-varma committed 3 years ago

3529a48e

optimize channels last for BatchNorm2d on CPU (#59286)

mingfeima committed 3 years ago

f3218568

[quant][fx] Remove extra q-dq for weight bias in normalization ops (#59882)

supriyar committed 3 years ago

864d129b

Build an -Wextra around c10 (#59853)

r-barnes committed 3 years ago

60eb22e4

make --run-specified-test-case use --include (#59704)

walterddr committed 3 years ago

e41bc31e

Fix some issues in CUDACachingAllocator (#59819)

r-barnes committed 3 years ago

cf0c4ac2

[nnc] Added a check to vectorize only those loops that are normalized. (#59423)

navahgar committed 3 years ago

b83ac0cc

Newer Older