strisunshinewentingwang/pytorch

Pull Requests Commits

[optim][adadelta] default to foreach when CUDA + differentiable=False

janeyx99 committed 1 year ago

e3e7a3e6

[ONNX] Documentation for `torch.onnx.find_mismatch` (#90728)

BowenBao committed 1 year ago

c537f5be

[utils][foreach] Add group tensor by device and dtype util (#92014)

janeyx99 committed 1 year ago

ed7885c2

[Inductor] Added aten.uniform_ decomp (#90869)

min-jean-cho committed 1 year ago

af242eed

[Dynamo] Fix guard bug when np.float used in control flow (#91991)

yanboliang committed 1 year ago

f40777e4

Python Script Object to IValue (#91776)

YazhiGao committed 1 year ago

8007c2d9

Add utility report_compile_source_on_error (#91069)

ezyang committed 1 year ago

8b00c544

In inductor triton generated code, avoid masking when numel=1 (#91254)

Fabio Rocha committed 1 year ago

4e21fc20

Not explicitly set the manifest filename in Windows (#91988)

huydhn committed 1 year ago

33e3c9ac

Update _optim_utils.py (#91935)

rohan-varma committed 1 year ago

a155f649

Third batch of canonical aten ops (#91995)

SherlockNoMad committed 1 year ago

28c736a4

Make ModuleList derive from Sequence[T] and type it appropriately (#89135)

aslvrstn committed 1 year ago

d0bfd79f

Revert "optimize sampled_addmm performance on CPU (SparseCSR) (#90978)"

pytorchmergebot committed 1 year ago

c5836153

Support --dynamic-ci-skips (#91893)

ezyang committed 1 year ago

74cbf058

Disable NVFuser in internal (Meta) build (#91836)

davidberard98 committed 1 year ago

83e6e9dd

Remove DL_RUNTIME_BUG (#91960)

peterbell10 committed 1 year ago

4806a9e7

[static-runtime] clamp fast_sigmoid result into (0,1) range (#91993)

tenpercent committed 1 year ago

6287bb78

[modes] make python arg parser also check for python key (#91573)

samdow committed 1 year ago

d8e795ec

[Modes] Add assert that the mode isn't already on the stack (#90770)

samdow committed 1 year ago

70283863

[stateless] add weight tying support (#90477)

samdow committed 1 year ago

8b3c4bc4

Add bf16 and change header file include path (#91838)

guangyey committed 1 year ago

e03ac0ee

s/INDCUTOR/INDUCTOR/ (#91885)

ezyang committed 1 year ago

d24324bf

Preventing crashing incase of no network by loading from cache (#91569)

MohammedZ666 committed 1 year ago

84b819d0

enable `conj()` for sparse compressed tensors (#91695)

nikitaved committed 1 year ago

850cf894

hrnet_w18, tts_angular works with dynamic shapes (#91891)

ezyang committed 1 year ago

56ed976e

Support zero dimensions in softmax decompositions (#91322)

davidberard98 committed 1 year ago

d7dc1c2f

replace vec::vec_scalar_t with at::opmath_type (#91086)

jiayisunx committed 1 year ago

afd8dd08

inductor: fix .to(memort_format) issue which doesn't generate right stride (#91948)

XiaobingSuper committed 1 year ago

3790b505

[SDPA] Guard mem efficient attention in deterministic mode (#91979)

drisspg committed 1 year ago

92855a21

[ONNX] Fix 'prim::PackPadded' shape inference (#91829)

BowenBao committed 1 year ago

d540442e

Older