strisunshinewentingwang/pytorch

Pull Requests Commits

[inductor] Parallelize Max Autotune step 2: Use multiple GPUs (#109127)

masnesral committed 1 year ago

4a09ed54

[inductor] Parallelize Max Autotune step 1: refactor autotune_process (#109126)

masnesral committed 1 year ago

ce428393

Remove include_0d from sample_inputs_gather (#109125)

guilhermeleobas committed 1 year ago

dbddf181

Update take_along_dim docs to include `dim=None` case (#109120)

guilhermeleobas committed 1 year ago

61f05787

Dispatch `numpy.take_along_axis` to `torch.take_along_dim` (#108880)

guilhermeleobas committed 1 year ago

d046376c

Add `SymInt` support to `torch.take_along_dim` (#108879)

guilhermeleobas committed 1 year ago

49e3d766

Fix failing inductor test (#109220)

drisspg committed 1 year ago

aca3bd44

Better error handling for cond (#108817)

ydwu4 committed 1 year ago

33c94b8b

Revert "add Half support for BatchNorm on CPU (#102070)"

pytorchmergebot committed 1 year ago

04a765f9

Disable tests mentioned in 109213 (#109232)

clee2000 committed 1 year ago

c44f8169

[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

Newer Older