strisunshinewentingwang/pytorch

Pull Requests Commits

tweak heuristic for sdpa selection based off of *data* (and a decision tree) (#99644)

Chillee committed 1 year ago

547bef11

Remove extra space (#99750)

ezyang committed 1 year ago

bb830224

[FSDP][BE] Remove unused code (#99731)

fegin committed 1 year ago

4f62e7cb

Fix decision logic for `should_cast_forward_inputs` in `_root_pre_forward()` and `_pre_forward()` (#99546)

speediedan committed 1 year ago

363d5300

Handle meta['val'] for tuple of lists. (#99724)

ezyang committed 1 year ago

10c938ab

Fix wrong path when reinstalling MacOS pip requirements (#99758)

huydhn committed 1 year ago

6c899999

[CI] Change max-autotune's output file name (#99754)

desertfire committed 1 year ago

db46d9dc

Improve OOM error message (#99699)

zdevito committed 1 year ago

8548cb3d

Disable XProtect on MacOS runner (#99506)

huydhn committed 1 year ago

c39aff10

Give distinct names to __unknown_tensor (#99729)

ezyang committed 1 year ago

18fd6394

Simplify _use_grad_for_differentiable (#98706)

jansel committed 1 year ago

b9da79d2

[Inductor] Fix rand_like with kwargs device of str type (#99673)

yanboliang committed 1 year ago

08376cc5

[FSDP][optim_state_dict] Consolidate rank0_only load logic (#99647)

fegin committed 1 year ago

7876c503

[FSDP][optim_state_dict] Support rank0_only when use_orig_params is on (#99624)

fegin committed 1 year ago

dd07dab1

Revert "Disable XProtect on MacOS runner (#99506)"

pytorchmergebot committed 1 year ago

fc63d710

[Feature] storage pin memory support custom device. (#99712)

wbigat committed 1 year ago

ee5f09ab

[stacktraces] Keep addr2line procs around (#99670)

zdevito committed 1 year ago

400075f7

[CI] Remove inductor skip list for Huggingface (#99375)

desertfire committed 1 year ago

e09f785a

Revert "[quant][pt2e][refactor] Cleanup the logic for deciding whether to insert observer/fq or not (#99220)"

pytorchmergebot committed 1 year ago

75e75480

[spmd] Introduce ParallelMode and add DTensorExpandMode (#98452)

wanchaol committed 1 year ago

b96bb2f1

[Inductor] Fix view/reshape on tensors with shape 0 in any dimension (#99671)

yanboliang committed 1 year ago

9244264f

[quant][pt2e][refactor] Cleanup the logic for deciding whether to insert observer/fq or not (#99220)

jerryzh168 committed 1 year ago

d56adb1b

Update docstring of torch.nn.functional.normalize() (#99512)

mega-optimus committed 1 year ago

06081ac8

Delete tracing_mode argument to export (#99555)

ezyang committed 1 year ago

e9786149

Move more stuff to after_aot (#99557)

ezyang committed 1 year ago

881c5723

Do not assume static by default when exporting (#99554)

ezyang committed 1 year ago

d3bb762f

[BE] Build PyTorch with `-Wnewline-eof` (#99687)

malfet committed 1 year ago

6b8ef8ea

Fix torch.nn.FractionalMaxPool2d output_size error (#99507)

tfsingh committed 1 year ago

dbf0db95

Revert "[c10d] Faster coalescing (#98793)"

pytorchmergebot committed 1 year ago

9861ec97

Revert "fix onednn ConvTranspose2d channels last issue when ic=1 (#99539)"

pytorchmergebot committed 1 year ago

da57d597

Newer Older