strisunshinewentingwang/pytorch

Pull Requests Commits

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

LucasLLC committed 1 year ago

f31d96d8

[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

`dl_open_guard` should restore flag even after exception (#96231) (#96457)

malfet committed 1 year ago

Verified ca0cdf52

[Release/2.0] Use Triton from PYPI (#96010)

malfet committed 1 year ago

Verified 9cfa076d

[Release/2.0] Use builder release branch for tests

malfet committed 1 year ago

8e05e41d

Remove mention of dynamo.optimize() in docs (#95802) (#96007)

msaroufim committed 1 year ago

Verified d8ffc60b

[MPS] Disallow reshape in slice (#95905) (#95978)

kulinseth committed 1 year ago

Verified 14837230

[MPS] Add fixes for div with floor (#95869)

kulinseth committed 1 year ago

Verified c4572aa1

[MPS] Fix views with 3 or more sliced dimensions (#95762) (#95871)

kulinseth committed 1 year ago

Verified 82b078ba

Remove torch._inductor.config.triton.convolution (#95840)

jansel committed 1 year ago

Verified 77f7bc5f

[optim] _actually_ default to foreach (#95862)

janeyx99 committed 1 year ago

Verified 08659645

Release version of fixed nll decomp (#95853)

ngimel committed 1 year ago

Verified f18ac1b3

[ROCM] Restrict pytorch rocm to only use triton 2.0.x (#95793) (#95834)

dllehr-amd committed 1 year ago

Verified c04134cd

[BE] Fix TORCH_WARN_ONCE (#95559) (#95822)

malfet committed 1 year ago

Verified 72d0863a

Update copyright (#95652) (#95700)

atalman committed 1 year ago

Verified 1bd334dc

[MPS] Remove FFT from the fallback as its causing crashes in test_ops and TestConsistency tests. (#95625)

kulinseth committed 1 year ago

Verified 93e13cd4

Older