strisunshinewentingwang/pytorch

Pull Requests Commits

aot autograd: handle detach() and no_grad() mutations on input (#95980)

bdhirsh committed 1 year ago

9d04fb9d

Remove rocm python 3.11 restriction (#99552)

atalman committed 1 year ago

Verified 661fa0c5

[release 2.0.1] [fix] fix load_sharded_optimizer_state_dict error on multi node (#99103)

eracah committed 1 year ago

Verified 0f49e97b

Fix flaky Dynamo export tests (#96488) (#99459)

peri044 committed 1 year ago

Verified b90fd012

Release branch cherry-pick of #96009 (#98717)

ngimel committed 1 year ago

Verified 3aae95a8

[Release only change] mitigate issue caused by deprecated pip option (#99487)

atalman committed 1 year ago

Verified 34dd578b

ngimel committed 1 year ago

Verified 654da19c

[MPS] Fix batch_norm_backwards key (#98794) (#99456)

malfet committed 1 year ago

Verified a470c041

Workaround for CuDNN-8.7+ load bug (#98644) (#99451)

malfet committed 1 year ago

Verified a943df04

Cherry-pick of #97214 (#98718)

ngimel committed 1 year ago

Verified bbf91554

Adds 3D attn_mask support to merge_masks() for Multihead Attention fast path (#98991) (#99092)

LucasLLC committed 1 year ago

Verified 3f25fc3b

[MPS] Fix the MacOS 13.3 selector check. (#98910)

kulinseth committed 1 year ago

Verified 763cfc21

[release 2.0.1] Warn once for TypedStorage deprecation (#98777)

kurtamohler committed 1 year ago

Verified ea9ea402

Revert "`GradScaler` recomputes `optimizer_state["found_inf_per_device"]` before `optimizer.step` (#97415) (#97886)" (#98624)

crcrpar committed 1 year ago

Verified 2e761498

Fix MPI rank and world size pg initialization (#98623)

H-Huang committed 1 year ago

Verified c8f24700

[cuDNN][cuDNN V8 API] Fix incorrect use of emplace in the benchmark cache (#97838) (#98528)

eqy committed 1 year ago

Verified f08d20ad

Forward fix for DataLoader to accept custom Sharding DataPipe (#97287) (#98523)

NivekT committed 1 year ago

Verified 94f88b34

Add support for custom backend (#95072) (#98513)

H-Huang committed 1 year ago

Verified 5c035a51

Clarify the saving of intermediates in the "extending torch.func" docs (#98020) (#98081)

zou3519 committed 1 year ago

Verified 8963d771

[release v2.0.1] Revisit `torch._six.string_classes` removal (#97737, #97789, #97863) (#98055)

XuehaiPan committed 1 year ago

Verified c039d2f5

[vmap] Fix index_select support when dim is negative (#97916) (#98019)

zou3519 committed 1 year ago

Verified 81ea036f

Fix module backward pre-hooks to actually update gradient (#97983) (#98017)

soulitzer committed 1 year ago

Verified c0c76df6

[fix] jacrev and jacfwd : support non-tensor args again (#97746) (#97908)

kshitij12345 committed 1 year ago

Verified cb51da82

`GradScaler` recomputes `optimizer_state["found_inf_per_device"]` before `optimizer.step` (#97415) (#97886)

crcrpar committed 1 year ago

Verified 57cd423a

Fix wrong handling of `grad_scale` & `found_inf` in fused optimizers (#95847) (#97885)

crcrpar committed 1 year ago

Verified 78e2cc6d

Fix type hint for `torch.Tensor.grad_fn` (#96804) (#97867)

XuehaiPan committed 1 year ago

Verified e9606385

cherry-pick: 1726c6f7a7723aa8222539db4e7747284c4b14de (#97857)

kshitij12345 committed 1 year ago

Verified e31e512f

[inductor] use triu ref instead of lowering (#96040) (#96462)

atalman committed 1 year ago

Verified c263bd43

Add jinja2 as mandatory dependency (#95691) (#96450)

atalman committed 1 year ago

Verified c9913cf6

Fix expired deprecation of comparison dtype for NumPy 1.24+ (#91517) (#96452)

atalman committed 1 year ago

Verified 2f7d8bbf

Older