strisunshinewentingwang/pytorch

Pull Requests Commits

Delete non-dynamic shapes export special case in guard creation (#103295)

ezyang committed 1 year ago

7be2a622

[Inductor][FX passes] Normalize torch.cat for pre_grad fusion (#102951)

yanboliang committed 1 year ago

1eb762c9

[DOCS][DDP]Fix the simple of saving and reloading PowerSGD state and hook. (#102721)

shaoyf42 committed 1 year ago

443edb90

[Dynamo] Support methods of NamedTuple (#103217)

fegin committed 1 year ago

fff5daf3

Properly respect automatic dynamic config for unspec int (#103321)

ezyang committed 1 year ago

d84b63c4

[pt2] add meta and `SymInt` support for `linalg_matrix_exp` (#102945)

nkaretnikov committed 1 year ago

2b3d955f

[pt2] bug fix: invert condition in `checkFloatingOrComplex` (#102944)

nkaretnikov committed 1 year ago

3a0f3773

[Reland2] fix missing-prototypes warnings in torch_cpu (Part 5) (#102931)

cyyever committed 1 year ago

34ccd1dd

Revert "Add distributed category to TORCH_LOGS (#103287)"

pytorchmergebot committed 1 year ago

90110b0e

[inductor] Support complex fallback for convert_element_type, _fft_c2c, view_as_real to support GoogleFnet with cpp wrapper (#103183)

davidberard98 committed 1 year ago

cde46572

[Pytorch] Add Vulkan support for aten::unsqueeze, 1d->2d, 3d->4d (#102987)

lucylq committed 1 year ago

f49b2f11

Revert "NCCL process group: avoid workEnqueue when capturing cuda graph (#102542)" (#103341)

ZainRizvi committed 1 year ago

89632b56

Add support for dictionary with torch object keys. (#103158)

ysiraichi committed 1 year ago

7550ec16

Revert "Make HigherOrderOperator stop appearing like torch.ops.* in FX (#103108)"

pytorchmergebot committed 1 year ago

d1f24f73

Add distributed category to TORCH_LOGS (#103287)

wconstab committed 1 year ago

0b252aeb

Revert "CUDA graphs overrides dynamic shapes and forces specialization (#103290)"

pytorchmergebot committed 1 year ago

d89dd05e

Fix calls to unqualified format_to to not clash with C++20's std::format_to (#103130)

ppwwyyxx committed 1 year ago

5aefa61d

NCCL process group: avoid workEnqueue when capturing cuda graph (#102542)

bottler committed 1 year ago

74a5d62d

Cleanup scatter-related code (#103074)

DamianSzwichtenberg committed 1 year ago

88aea179

CUDA graphs overrides dynamic shapes and forces specialization (#103290)

ezyang committed 1 year ago

c760f0e4

change pre_autograd to pre_dispatch tracing (#101818)

bdhirsh committed 1 year ago

b0392de2

Resolve TODO by deleting assert sparse cannot be meta on SymInt (#103299)

ezyang committed 1 year ago

1c3a7d9a

Record view stacks if running anomaly mode (#103185)

ezyang committed 1 year ago

a02c573a

Revert "Make torch.compile(dynamic=True) not assume static by default (#99469)"

pytorchmergebot committed 1 year ago

79e0a1ea

Remove capture_scalar_outputs sanity check prepping for dynamic by default (#103292)

ezyang committed 1 year ago

2e21cb09

Disable dynamo'd test_optim entirely (#103323)

janeyx99 committed 1 year ago

4a5d56b7

Add mmap option to `torch.load` (#102549)

mikaylagawarecki committed 1 year ago

6fa2d41d

Move tensor grouping to ATen (#100007)

crcrpar committed 1 year ago

74b7a6c7

Make torch.compile(dynamic=True) not assume static by default (#99469)

ezyang committed 1 year ago

7108c035

Preserve CreationMeta when metafying views. (#103152)

ezyang committed 1 year ago

96fd2836

Newer Older