strisunshinewentingwang/pytorch

Pull Requests Commits

[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

Fix names of some reference functions (#88115)

lezcano committed 1 year ago

ce0e22a8

Add platform markers for linux only extra_install_requires (#88826)

JacobHayes committed 1 year ago

2e358cc9

Export c10/[macros|util] headers to be used by internal inductor builds (#89249)

malfet committed 1 year ago

5654fed2

[PT-D][Checkpoint] Update import and update docstring for distributed checkpoint (#89256)

wz337 committed 1 year ago

4c672498

[LTC] Upstream short_metrics (#89186)

alanwaketan committed 1 year ago

2dcacc6b

[ao] maintain BC for is_activation_post_process (#89260)

HDCharles committed 1 year ago

c5fafb4e

Disable tracing `zero_grad()` (#88731)

mlazos committed 1 year ago

30c3e5af

Gate CUDA-only inductor tests by HAS_CUDA (#89251)

huydhn committed 1 year ago

afdc48f8

[flaky] relax tolerance conv1d_vs_scipy (#89193)

kshitij12345 committed 1 year ago

6a964c16

Newer Older