strisunshinewentingwang/pytorch

Pull Requests Commits

[caffe2][cuda] Fix instrumentation of malloc/free SDTs for `CUDACachingAllocator` (#108907)

vlad-scherbich committed 1 year ago

2d26364f

Fix issue when input/output buffer of functional collective (e.g. allreduce / allgather) is incorrectly reused later (#108811)

pytorchbot committed 1 year ago

faa5985d

[FSDP] Only check exec order if DETAIL (#109049)

awgu committed 1 year ago

54dd65f9

[MPS] Fix crash if nonzero is called concurrently (#108996)

malfet committed 1 year ago

916183a0

Do not use a specific LOC in link (#108957)

spzala committed 1 year ago

35aeb6aa

Improve type annotations for `jit.script` (#108782)

randolf-scholz committed 1 year ago

32f50b70

Back out "[Inductor] Extend Pattern Matcher to Match Equivalent Function Invocation (#107832)" (#109174)

hl475 committed 1 year ago

8851603a

[PyTorch] Add Expanded call stack to nodes (#108426)

RihamSelim committed 1 year ago

c657d9ec

Use global variables to register the return_types namedtuples (#108832)

andreigh committed 1 year ago

00908475

add Half support for BatchNorm on CPU (#102070)

CaoE committed 1 year ago

6065e7a9

Use the correct channel token when uploading nightly triton conda (#109073)

huydhn committed 1 year ago

f6d8ecf9

Allow marking multiple unstable configs of the same job name (#109185)

huydhn committed 1 year ago

c9fdfafb

inductor/test_max_autotune serial in CI (#109209)

clee2000 committed 1 year ago

fe198f31

Add missing DeviceMesh import (#109187)

huydhn committed 1 year ago

d05a6e5a

Back out "Dynamo support for autograd.Function w/ once_differentiable (#108686)" (#109199)

redraven984 committed 1 year ago

f2639a2c

[inductor] Enable Mypy Checking for torch/_inductor/codecache.py (#108789)

masnesral committed 1 year ago

264f1e7b

Flash Attention v2 (#105602)

drisspg committed 1 year ago

ad90ab31

optests improvements based on torchvision usage on nms (#108929)

ezyang committed 1 year ago

55f956f1

[optests] Changed failures_dict format to json; automatic update of failures_dict (#109110)

zou3519 committed 1 year ago

bfa8429c

Check index size during decomp of index_add (#108826)

int3 committed 1 year ago

db48bc80

Enable typechecking for _inductor/virtualized.py (#108916)

int3 committed 1 year ago

d2d36aad

Revert "[dynamo] preserve some FX node metadata of GraphModules (#107067)"

pytorchmergebot committed 1 year ago

c5e75886

torch/csrc/profiler/README.md - stubs, RecordFunction, Autograd interaction (#108470)

davidberard98 committed 1 year ago

aee5dec3

Use user directed names for variables where possible (#109092)

voznesenskym committed 1 year ago

de0b18fa

Forward fix lint (#109177)

clee2000 committed 1 year ago

015be4ce

Update inductor ci_expected_accuracy (#109148)

angelayi committed 1 year ago

3d8d59e6

Fix `torch._numpy.random` (#108944)

ev-br committed 1 year ago

3ac2396e

Symintify repeat_interleave (#109133)

anijain2305 committed 1 year ago

41e5d410

Add torch.export.register_dataclass API (#109152)

gmagogsfm committed 1 year ago

a09539f4

[dtensor][4/n] don't use make_fx for strategy propagation (#108262)

wanchaol committed 1 year ago

375d2ca6

Newer Older