strisunshinewentingwang/pytorch

Pull Requests Commits

Update sdp dispatch logic to enable fused backward (#89154)

drisspg committed 1 year ago

2e72ec79

[dynamo] mutable local caching to make dynamo faster at tracing mutation (#89170)

mlazos committed 1 year ago

85a87e63

Move bazel to c++17 (#89297)

malfet committed 1 year ago

ea58955d

[dashboard][huggingface] skip accuracy checks for really large models… (#89273)

anijain2305 committed 1 year ago

cad5772c

[small] Update error message (#89294)

H-Huang committed 1 year ago

ee907375

[functorch] introduce an experimental map() op. (#88767)

zhxchen17 committed 1 year ago

c3938bb9

Detach fake tensors into val, so they aren't affected by metadata mutation (#89140)

ezyang committed 1 year ago

94b5c807

[BE] Print backtraces from coredumps (#89309)

malfet committed 1 year ago

885f8a56

[FX] Add type annotation to `getitem` node before `split_module` (#88510)

lequytra committed 1 year ago

0e1fcc8a

[Inductor CI] Use string format for cuda-arch-list input to prevent 8.0/9.0/10.0 etc from being interpreted as 8/9/10 (#89279)

weiwangmeta committed 1 year ago

ecfb4e06

Add NVTX markers that dump additional information for nvprim_nvfuser Dynamo graphs (#88259)

blongnv committed 1 year ago

7551136b

[Profiler] Don't raise SOFT_ASSERT in debug builds. (#89240)

Taylor Robie committed 1 year ago

35d5fc52

[DDP][Docs] Add warning that `no_sync()` should include forward (#89244)

awgu committed 1 year ago

bfffc8d8

Re-enable test_hf_bert_fsdp (#89223)

davidberard98 committed 1 year ago

304b5de1

Don't trace when we track_tensor_tree (#89139)

ezyang committed 1 year ago

ba605c3b

Symintify obeys_layout_contract (#89138)

ezyang committed 1 year ago

e04dc35a

Remove --retry-all-errors from environment with old curl (#89298)

ZainRizvi committed 1 year ago

837ca8f3

Set make max load when building libtorch (#89237)

huydhn committed 1 year ago

ee2ce3fe

Vectorized horizontal flip implementation (#88989)

vfdev-5 committed 1 year ago

7ec8a4d2

[Dynamo] Support Tensor.nelement & torch.cuda.is_available (#89164)

yanboliang committed 1 year ago

81a4aeab

Added partial decomposition of conv_backward and grad_bias computation (#89128)

Chillee committed 1 year ago

8a419cbf

[quant][fx][be] Refactor replace observer with q/dq op code (#89247)

jerryzh168 committed 1 year ago

38ccd08f

Use standard __func__ macro in symbolic shape. (#89264)

zhxchen17 committed 1 year ago

c219b55b

[xplat] remove -weak_framework (#89233)

rmaz committed 1 year ago

12a97444

[Quant] Allow setting fixed qparams for inner LSTM ops (#88456)

andrewor14 committed 1 year ago

19e66fce

[inductor] Skip DALLE2_pytorch in torchbench (#89288)

desertfire committed 1 year ago

19fcb805

[inductor] Temporarily disable functorch_dp_cifar10 test in TorchBench (#89281)

desertfire committed 1 year ago

1f7c0ff6

Update torch.distributed.DistBackendError type (#89235)

H-Huang committed 1 year ago

55e55d95

Add most in-place references/decompositions (#88117)

lezcano committed 1 year ago

154e58c0

Simplify maybe_resize_out (#88116)

lezcano committed 1 year ago

6741443c

Newer Older