strisunshinewentingwang/pytorch

Pull Requests Commits

Don't build Triton from source in benchmarks/dynamo/Makefile (#100613)

ezyang committed 1 year ago

c58d9642

[docs] Docs for writing ATen IR passes + FX Pattern matching (#100577)

angelayi committed 1 year ago

8eb82135

Add AotAutogradFallbackTests to dynamic suite (#100454)

voznesenskym committed 1 year ago

fe3ecfe0

Reland basic dynamo support for traceable collectives (#100476)

wconstab committed 1 year ago

2dca4181

Fix graph break in a common `func(self, *args)` pattern (Faster stable diffusion) (#100444)

voznesenskym committed 1 year ago

9f3c6b1b

Improve minifier printing to be more chatty when it makes sense (#100486)

ezyang committed 1 year ago

c2556c03

Misc accuracy improvements on minifier (#100447)

ezyang committed 1 year ago

c7e9f406

[codemod] Replace hasattr with getattr in caffe2/test/distributed/fsdp/test_fsdp_optim_state.py (#100360)

r-barnes committed 1 year ago

7f997aa3

[vision hash update] update the pinned vision hash (#100510)

pytorchmergebot committed 1 year ago

8df748f3

Fix bug in process_group_name when there is duplicate pgs (#100518)

xw285cornell committed 1 year ago

c29ab841

[replicate] input casting support (#100216)

rohan-varma committed 1 year ago

253b9d32

[primTorch] add ref for `polar` (#100345)

nkaretnikov committed 1 year ago

e87ed2a8

Ensure device arg is passed to test_transformers (#100260)

drisspg committed 1 year ago

2892c06e

inplace PyTorchStreamReader getRecord() (#100418)

jiayisuse committed 1 year ago

f558bb6f

Use Boxed Calling Convention for AOT Eager (#100417)

eellison committed 1 year ago

e6c0164f

Require contiguous for view_as_complex (#100428)

eellison committed 1 year ago

d67e4db8

Remove speech_transformer workaround, torchbench handles it correctly now (#100558)

ezyang committed 1 year ago

d25c93f9

[dynamo] Minor fixes and clean-up in eval_frame.c (#100496)

colesbury committed 1 year ago

fd841763

add checkpoint support for custom device (#99626)

heidongxianhua committed 1 year ago

6aeb85ad

[cuDNN][CUDA] Fix for `install_cudnn.sh` following 12.1 CI update (#100501)

eqy committed 1 year ago

3c1dd0a4

[C10D] Drop the GIL when creating a TCPSTore to avoid deadlocks. (#100555)

Rodrigo Kumpera committed 1 year ago

fa2bfab9

Support multiple separator characters, '/' and '\\', on Windows. (#98146)

masamitsu-murase committed 1 year ago

c3bcf5f6

[ONNX] Update 'Functionalize' pass to support pre-decomp graph; Drop 'aten_graph' arg for 'DynamoExporter' (#99667)

BowenBao committed 1 year ago

f8275633

[pytorch] Accelerate indexing_backward_kernel with duplicates (#99441 attempt 2) (#100505)

valentinandrei committed 1 year ago

9bc68fcd

[reland][CI] Start to collect inference perf with cpp_wrapper ON (#100187) (#100502)

desertfire committed 1 year ago

61813a8e

Check uppercase when checking for merge blocking SEVs (#100559)

kit1980 committed 1 year ago

1a6f613b

[MPS] Add dot input check (#100099)

qqaatw committed 1 year ago

0a6a0ac4

Partition modules (#98628)

angelayi committed 1 year ago

3c5ec6af

Properly propagates checkpoint wrapper args and kwargs (#99791)

dirkgr committed 1 year ago

75945d54

[cuDNN][cuDNN V8 frontend API] Clean up `time_sorted_plan` workaround for cuDNN v8 API (#100287)

eqy committed 1 year ago

8f6951cf

Newer Older