strisunshinewentingwang/pytorch

Pull Requests Commits

[pt2] Register meta func to randperm.default (#99593)

xw285cornell committed 1 year ago

cc01568e

[ROCM] Do not send "debug"=True down to triton.compile (#99756)

dllehr-amd committed 1 year ago

8a0badff

Skip some failing dynamic shape models on periodic (#99895)

huydhn committed 1 year ago

9a69634b

inductor: fix issue for conv+binary issue for binary scalar path (#99860)

XiaobingSuper committed 1 year ago

df1ff092

inductor: add fallback test case for hardtanh and leakyrelu fusion pattern (#99859)

XiaobingSuper committed 1 year ago

ed395779

Fix typos and clarify text in tags.yaml (#99954)

ezyang committed 1 year ago

e2cb6bcc

[ONNX] Bump onnx-script version with imported module renaming (#99926)

titaiwangms committed 1 year ago

4cf65462

Return full accuracy status for printing, not abbreviated version (#99894)

ezyang committed 1 year ago

04e8df4d

div16 changes for dyn shapes (#99930)

voznesenskym committed 1 year ago

bfa63bf4

[dynamo] avoid graph break on repeat_interleave.self_int (#99528)

jgong5 committed 1 year ago

e5c9a0fc

Implement the get_device method in the storage base class. (#99818)

fakeYan committed 1 year ago

ecd2c718

[ONNX] Improve diagnostics performance (#99936)

BowenBao committed 1 year ago

e5145329

Add fsspec to PT setup.py (#99768)

wz337 committed 1 year ago

466adab7

Allow persistent reductions in dynamic shapes if last numel is static (#99789)

voznesenskym committed 1 year ago

4be0aa13

yolov3 dynamic training accuracy is fixed (#99896)

ezyang committed 1 year ago

cd617071

inductor: align inductor behavior with eager mode for split_with_sizes (#99702)

XiaobingSuper committed 1 year ago

41069f2f

Use memoized only mode for guard size/stride extraction (#99742)

voznesenskym committed 1 year ago

96ceae3a

Stop marking sequence length as dynamic (#99889)

ezyang committed 1 year ago

0b545bc6

[torchgen] accept scalars for unary `SymInt` arrays (#99921)

nkaretnikov committed 1 year ago

42921fc8

make ATen/native/cuda/NaiveConvolutionTranspose3d.cu data_ptr-correct (#99347)

mikey dagitses committed 1 year ago

1dbecbf9

make ATen/native/cuda/SortStable.cu (#99340)

mikey dagitses committed 1 year ago

4ca44d32

make ATen/native/cuda/RreluWithNoise.cu (#99341)

mikey dagitses committed 1 year ago

1b30f588

[pt2] add `SymInt` support for trapezoid ops (#99281)

nkaretnikov committed 1 year ago

fbb0ff10

De-select odd numbered heads from nn.MHA fastpath (#99672)

mikekgfb committed 1 year ago

36e1ae67

[FSDP][Reland] Include duplicate parameters and modules when calling named_parameters and named_modules (#99448)

fegin committed 1 year ago

3de7fd46

Change export tracing_mode default to symbolic (#99877)

ezyang committed 1 year ago

0eb59ad0

Do not assume static by default when exporting (#99554) (#99876)

ezyang committed 1 year ago

73f7459a

[FSDP] Set `NCCL_DESYNC_DEBUG=0` for FSDP unit tests (#99916)

awgu committed 1 year ago

08a8a37f

[spmd] skip gradient copying for fused adam (#99489)

wanchaol committed 1 year ago

855f611b

minor fix to fused adam meta registration (#99436)

wanchaol committed 1 year ago

ca24a962

Newer Older