strisunshinewentingwang/pytorch

Pull Requests Commits

Fix TODOs related to #38095 in test_mps.py (#89815)

kit1980 committed 3 years ago

09f2373e

Revert "Call _sdp_attention in nn.functional.mha (#89470)"

pytorchmergebot committed 3 years ago

f1415b8c

Revert "replace double transpose with single permute in nn.f.mha (#89847)"

pytorchmergebot committed 3 years ago

618a585f

Add a cpp wrapper for Inductor (#88167)

chunyuan-w committed 3 years ago

a6caa9c5

[FSDP][Easy] Remove internal default arg (#89227)

Andrew Gu committed 3 years ago

5949d5fe

add bf16 in fp32 out fast path for embedingbag in caffe2 perfkernel (#89198)

zhuhaozhe committed 3 years ago

7cd6e6ac

[benchmarks][dynamo] Trying CI - Set train() for TIMM models accuracy tests (#89780)

anijain2305 committed 3 years ago

68805b08

Revert "[aarch64] add SLEEF dependency for aten_cpu (#89475)"

pytorchmergebot committed 3 years ago

969a7d09

Revert "Add bits tensor types (#88594)"

pytorchmergebot committed 3 years ago

4cc5be3a

Row and column select support for block compressed sparse tensors (#88733)

pearu committed 3 years ago

296e1ba4

[PT-D][Checkpoint]Add MultiThreaded FileSystemWriter for distributed checkpointing and Update tests (#87987)

wz337 committed 3 years ago

0cc0e5ef

fix RowwiseMoments vectorization issue on CPU (#84404)

mingfeima committed 3 years ago

87d18cf0

Vectorize erf (#89837)

EikanWang committed 3 years ago

92f08f09

[PT-D][Tensor Parallel] Add more test cases when we use use_orig_params for FSDP wrapping (#89779)

fduwjj committed 3 years ago

009dd3c4

Dynamo, FX, Inductor Progress Bars (#88384)

msaroufim committed 3 years ago

011452a2

[Dynamo] Fix source/reconstruction bugs in NNModule named_* calls (#89729)

yanboliang committed 3 years ago

d88b5555

Update DDP docs for Dynamo/DDPOptimizer (#89096)

wconstab committed 3 years ago

44728375

[dtensor] update README (#89800)

wanchaol committed 3 years ago

12f98f85

update subscriber list (#89799)

wanchaol committed 3 years ago

b09efae3

Add arguments to collect_results (#89611)

williamwen42 committed 3 years ago

f4707ae0

[FSDP] Include module classes in `ModuleWrapPolicy.__repr__` (#89058)

Andrew Gu committed 3 years ago

ce17bb95

[FSDP] Limit all gather after pre-unshard (#89057)

Andrew Gu committed 3 years ago

c8aaad04

[Checkpoint][2D][5/N] Add checkpoint_utils for distributed checkpoint to testing/_internal/distributed/ (#89873)

wz337 committed 3 years ago

56b3ad78

[FSDP] Remove unneeded stream sync from `clip_grad_norm_()` (#89308)

Andrew Gu committed 3 years ago

be80b72a

Generator of tensor inputs with variable layout and structure (batch/non-batch, hybrid/non-hybrid, block/non-block) (#88914)

pearu committed 3 years ago

90bed887

Enable rsqrt (#89771)

eellison committed 3 years ago

275ade63

add env/config flag to disable dynamo (#89828)

mlazos committed 3 years ago

2d32e5dd

[functorch] Move `cond.py` to `_cond.py` and expose `cond()` under functorch.experimental.control_flow. (#89819)

zhxchen17 committed 3 years ago

a70082a8

[FSDP][Easy] Remove outdated TODO (#89217)

Andrew Gu committed 3 years ago

d1760d7a

Make fake tensors preserve dense strides in type conversion (#89803)

eellison committed 3 years ago

1a33b7cb

Older