strisunshinewentingwang/pytorch

Pull Requests Commits

[FSDP] Provide a utility API to allow users easily to set state_dict_type (#73716)

fegin committed 2 years ago

be535e24

Back out "Revert D34524207: [pytorch][PR] remove _s_where" (#73579)

ngimel committed 2 years ago

e47a5a64

Fix segfault while real and imaginary attributes are set to a number (#73867)

anjali411 committed 2 years ago

37e0d2e3

[ROCM] Navi21 Enablement 5: Softmax kernels (#73545)

micmelesse committed 2 years ago

5dfbe52a

[codemod][type-comments] Convert type comments in examples.py (#73085)

stroxler committed 2 years ago

299dec1c

Add logging for ProcessGroup backends. (#73702)

pritamdamania committed 2 years ago

3b30c8da

Fix nightly docker publish build

janeyx99 committed 2 years ago

f982d6a6

[JIT] log extract tool - dump NVFuser fallbacks instead of fusion groups (#73881)

davidberard98 committed 2 years ago

31b64fc3

Fix libtorch_cuda_linalg builds (#73896)

malfet committed 2 years ago

Verified 56164c07

[ZeRO][BE] Clean up ZeRO tests (#73842)

awgu committed 2 years ago

9012e8d6

[FSDP] Generalize fsdp_modules() (#73553)

fegin committed 2 years ago

5372dcd9

[Qunat] Refactor reference module mapping (#72755)

terrychenism committed 2 years ago

4e6aefaf

[Model Averaging] Add a reference to hierarchical SGD (#73823)

wayi1 committed 2 years ago

5993f487

[FSDP] Add always_wrap policy (#73687)

rohan-varma committed 2 years ago

270c27ef

Fix docstring hiding due to #45689 (#73747)

fritzo committed 2 years ago

076619d7

record_function: add torchbind alternative API (#72301)

peterbell10 committed 2 years ago

9ef5c679

[fx/graph_drawer] Add skip_node_names_in_args option, default to True (#73815)

jfix71 committed 2 years ago

e99e3fa5

[SR] Make sigrid_transforms fusion work on graph outputs (#73091)

Mike Iovine committed 2 years ago

2c350960

Revert "ci: Migrate windows conda to GHA"

pytorchmergebot committed 2 years ago

be4bcf8f

add tools/onnx to merge rules

garymm committed 2 years ago

a3cff181

Deduplicate legacy _ctor and _new Python bindings (#73822)

ezyang committed 2 years ago

d6c29b1d

Add forward AD support for logsumexp, log_softmax, softmax, nll_loss, and cross_entropy (#73741)

soulitzer committed 2 years ago

de73f9a5

[RELAND] [cuDNN] Add a new optimized cuDNN RNN algorithm for small RNN hidden_size (#73211)

mcarilli committed 2 years ago

943085fe

Update __torch_dispatch__ to return op overload instead of the opoverload packet function (#72673)

anjali411 committed 2 years ago

086645ad

[JIT] make RegisterCudaFuseGraph use TORCH_API instead of C10_EXPORT (#73742)

davidberard98 committed 2 years ago

a3d099ea

[ONNX] Fix repeat interleave when repeats and dim is 1

BowenBao committed 2 years ago

6c26bf0e

[FSDP][BE] Change assert to assertEqual (#73787)

awgu committed 2 years ago

15a799e3

[Easy][FSDP] Fix warning render (#73786)

awgu committed 2 years ago

89b16f1a

[FSDP] Add grad accumulation without `no_sync()` (#73535)

awgu committed 2 years ago

4a06b8d3

[Static Runtime] Add out variant wrapper for aten::ones (#73851)

d1jang committed 2 years ago

71961d37

Newer Older