strisunshinewentingwang/pytorch

Pull Requests Commits

Refactoring to share vectorization code for int8/uint8. (#89650)

alexsamardzic committed 2 years ago

a78467f3

[minifier] Continue on assertion for accuracy minification (#89739)

anijain2305 committed 2 years ago

8226a5d3

[dynamo] Don't copy the graph during checkpointing (copy_graphstate) (#89232)

mlazos committed 2 years ago

40dd03ee

add memory_tracker tool to help profiling memory usages (#88825)

zhaojuanmao committed 2 years ago

91899a9e

Test FSDP with submodule non-reentrant checkpointing (#89781)

mrshenli committed 2 years ago

7ec7a820

Dynamo asserts FSDP wrapped modules use_orig_param (#89523)

wconstab committed 2 years ago

705ad36c

Enable DDPOptimizer by default in dynamo (#88523)

wconstab committed 2 years ago

7860fcc2

Move Dynamo docs back to core (#89769)

msaroufim committed 2 years ago

9048cf16

[dynamo] Minifier fixes for reproducing segfault (#89712)

anijain2305 committed 2 years ago

2b522670

[decomp] Fix native_batch_norm_backward dtype of dweight and dbias (#89740)

anijain2305 committed 2 years ago

c1950620

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

drisspg committed 2 years ago

4d7ec302

fixes for inductor <> batch norm (#89603)

bdhirsh committed 2 years ago

e20ec445

Add type hint to torch.norm and Tensor.norm (#89728)

fd0r committed 2 years ago

740860d4

[nvfuser] avoid out of bounds error (#89584)

davidberard98 committed 2 years ago

908daa8a

Special-case fsdp wrapped modules to be Unspecialized (#89330)

wconstab committed 2 years ago

77df2ca9

[Inductor] Add an option to mark wrapper call in PyTorch profiler (#89674)

jgong5 committed 2 years ago

c75434ed

[functorch] fix possible overflow (#83389)

cyyever committed 2 years ago

4b11119c

Fix archive issue impacting summary stat diff (#89789)

williamwen42 committed 2 years ago

63843401

[FSDP] Fix `nn.Parameter` usage for 2D and `use_orig_params=True` (#89782)

awgu committed 2 years ago

943acd4d

[Checkpoint][2D][4/N] Add nested_dict for distributed checkpoint to core distributed (#89537)

wz337 committed 2 years ago

23ee6757

Re-enabled 3 reductions tests on Windows (#89567)

mantaionut committed 2 years ago

a378ba21

Add bits tensor types (#88594)

angelayi committed 2 years ago

f3b1315e

[Checkpoint][2D][3/N] Add nested_tensors for distributed checkpoint to core distributed (#89501)

wz337 committed 2 years ago

22e7514a

[CUDA graphs] Add warning if captured graph is empty (#88754)

Aidyn-A committed 2 years ago

0057be33

[skip ci] documentation update for the kwargs defaults section of fun… (#89719)

weberxie committed 2 years ago

c18da597

[LTC] Metrics can be reset too (#89606)

alanwaketan committed 2 years ago

13d2af2a

[Vulkan][TCC] Fix conv2d pack biases (#89568)

manuelcandales committed 2 years ago

5abe454d

Add meta implementation for _efficientzerotensor (#88936)

peterbell10 committed 2 years ago

2e0cd7c8

Fix comparison of batched_prop vs unbatched_prob in test_distributions (#87977)

Flamefire committed 2 years ago

69a8c92d

Revert "Move Dynamo docs back to core (#89769)"

pytorchmergebot committed 2 years ago

47cca5e4

Newer Older