strisunshinewentingwang/pytorch

Pull Requests Commits

Disallow registering meta function for CompositeImplicitAutograd ops (#90222)

SherlockNoMad committed 1 year ago

42705bd7

pad low precision matmuls when requested (#90235)

Natalia Gimelshein committed 1 year ago

a88400e0

as_strided: Fix default storage_offset for reference implementation (#89513)

peterbell10 committed 1 year ago

ba70a8be

Functionalization: skip meta block computation if compute_reference_meta is false (#90219)

isdanni committed 1 year ago

05ccbd6d

Assert there are no outstanding side effects before calling cond (#90208)

ezyang committed 1 year ago

962ebe88

Revert "[Composable API] `replicate`: change to per module call, remove `mark_root_module()` (#89222)"

pytorchmergebot committed 1 year ago

0d8e53df

[vision hash update] update the pinned vision hash (#90239)

pytorchmergebot committed 1 year ago

73565ce3

Revert "[Composable API] `replicate`: add support for DDP args (#89243)"

pytorchmergebot committed 1 year ago

3749b9dc

TorchDynamo: always convert flexiblelayout to be FixedLayout when given a stride_order (#89904)

XiaobingSuper committed 1 year ago

2597d5d7

[inductor] Add test_ops_gradients running with inductor (#89792)

desertfire committed 1 year ago

29233a18

Dynamo FX graph stack traceback fix (#87136)

williamwen42 committed 1 year ago

ebeecbf8

Fix yet another C++17 Windows build issue (#90228)

malfet committed 1 year ago

a268b9e5

[Pytorch][Vulkan] Use specalized shader for 3x3 depthwise conv (#89953)

kimishpatel committed 1 year ago

55b10e6b

[Pytorch][Vulkan] Templatize depth wise convolution and specialize for 3x3 and

kimishpatel committed 1 year ago

a17765a1

[Pytorch][Vulkan] shader codegen use ordered dictionary (#89951)

kimishpatel committed 1 year ago

bd456fb5

[Pytorch][vulkan] Simplify depthwise conv to remove bounds compute (#89950)

kimishpatel committed 1 year ago

cb68dcbd

[Vulkan] output benchmark numbers for aibench parsing (#89949)

kimishpatel committed 1 year ago

876b7024

[pytorch][vulkan] realistic benchmark size for depthwise (#89948)

kimishpatel committed 1 year ago

841eba63

[Caffe2] Fix the assert message (#89816)

atuljangra committed 1 year ago

564905c8

Fix XLA dynamo CI (#90229)

JackCaoG committed 1 year ago

2ea32f41

Add sharding strategy to fully_shard (#90192)

mrshenli committed 1 year ago

5d6aa99c

Add a repro for fully_shard _unshard error (#90190)

mrshenli committed 1 year ago

e4670885

[Composable API] `replicate`: add support for DDP args (#89243)

yhcharles committed 1 year ago

0f274ed3

[FSDP][optim_state_dict][1/N] Restructure _optim_state_dict to prepare the support of use_orig_param (#89898)

fegin committed 1 year ago

72fdfad4

Simplify by using yield from (#90160)

kit1980 committed 1 year ago

2b20a3d3

Fix issue 38095 TODOs in NCCL tests (#90033)

kit1980 committed 1 year ago

54858cce

[NNC] Use New PassManager for LLVM >= 15 (#89978)

davidberard98 committed 1 year ago

7571134f

Assume that co_firstlineno is always defined (#90180)

ezyang committed 1 year ago

5de5c5e4

workaround for indexing formulas with negative terms (#89933)

Natalia Gimelshein committed 1 year ago

1ea20cdb

fix c10::detail::integer_iterator for C++17 (#90174)

mikey dagitses committed 1 year ago

368a1cbd

Newer Older