strisunshinewentingwang/pytorch

Pull Requests Commits

[FSDP][state_dict][5/N] Remove the FSDP module dependency from _state_dict_utils (#88637)

fegin committed 1 year ago

7aa144ac

Fix CUDNN_PATH handling on Windows (#88898)

malfet committed 1 year ago

575e02df

[fix] allow saving python attr on Tensor and Parameter via torch.save (#81616)

kshitij12345 committed 1 year ago

f7494632

Revert "rename DisableTorchFunction to DisableTorchFunctionSubclass (#88218)"

pytorchmergebot committed 1 year ago

ba4d5aae

Revert "add DisableTorchFunction that matches DisableTorchDispatch (#88219)"

pytorchmergebot committed 1 year ago

4e5d7afe

[ONNX] Add stack info to diagnostics (#87258)

BowenBao committed 1 year ago

9d7d21f5

[FSDP][state_dict][4/N] Move the core logic of summon full parameters to _unshard_params_utils.py (#88636)

fegin committed 1 year ago

3d1c5c89

Fix ATen Fallback for BUILD_CAFFE2=0 for ONNX-only ops (#88504)

Thiago Crepaldi committed 1 year ago

5f0783bd

Take input striding for conv forward based on eager output (#88706)

eellison committed 1 year ago

8ff2e34c

Revert "[Autograd] Use in-place input accumulation fast path for dense Tensors. (#88339)"

pytorchmergebot committed 1 year ago

adfbd831

Explicitly check filelike arg of `torch.save` (#88867)

kurtamohler committed 1 year ago

89a326ff

Regularize bernouilli_ with bernouilli decomp (#88349)

eellison committed 1 year ago

a6832b08

Symintify `broadcast_to` (#88776)

nkaretnikov committed 1 year ago

1e8f95ac

Add meta impl for topk (#88694)

anjali411 committed 1 year ago

d615d122

[FSDP][state_dict][3/N] Change how state_dict utils access attributes in _FSDPState (#88635)

fegin committed 1 year ago

3c7f9666

Add context manager to allow mutation on saved tensors (#79056)

soulitzer committed 1 year ago

b92acee8

[dynamo] Add torch.device to is_safe_constant (#88766)

desertfire committed 1 year ago

91b71cdb

[FSDP][state_dict][2/N] Move state_dict related enums/dataclasses/states to state_dict_utils.py, api.py and init_state_dict() (#88481)

fegin committed 1 year ago

324ac93a

Fix cuda/cpu check on NoneType (#88854)

mikekgfb committed 1 year ago

ee91c328

Error on ZeroTensor serialization (#88803)

kshitij12345 committed 1 year ago

d15a6b0c

[ONNX] Add test case for onnx::Max scalar type (#88751)

titaiwangms committed 1 year ago

b843f4db

Use `atomicAdd` for `bfloat16` in Ampere and above (#84981)

eqy committed 1 year ago

396c3b1d

[ONNX] Add onnx::Max into standard Op for scalar type alignment (#88750)

titaiwangms committed 1 year ago

a6d72f44

Revert "[dynamo] fixes dict changed during runtime error (#87526)"

pytorchmergebot committed 1 year ago

0de8f047

Update lr_scheduler.pyi to match lr_scheduler.py (#88818)

janeyx99 committed 1 year ago

310335de

Fix FakeTensorProp on Module with Parameters or Buffers (#88700)

wschin committed 1 year ago

86b7aa26

[FSDP][state_dict][1/N] Moving state_dict logic to pre_state_dict_hook (#87900)

fegin committed 1 year ago

c4fc5d37

Disable check for dropout in MultiheadAttention fast_path (#88831)

ecly committed 1 year ago

9d09968b

[vision hash update] update the pinned vision hash (#88853)

pytorchmergebot committed 1 year ago

30823787

Ref for aten.full; symint changes in prim (#88762)

SherlockNoMad committed 1 year ago

495e7b1c

Newer Older