strisunshinewentingwang/pytorch

Pull Requests Commits

extend serialization for tensor metadata (#99808)

fakeYan committed 1 year ago

b7777c81

Using dynamic allocation buffer and dynamic threads on scan with index (#103502)

mfkasim1 committed 1 year ago

ce0a5119

Make DDPOptimizer handle subgraphs without outputs (#103488)

wconstab committed 1 year ago

fee01640

Use CUDA DSA in ATen (#95300)

r-barnes committed 1 year ago

93b0410e

Checking for nullptr in get_model_bytecode_version (#97149)

mmigdal-nv committed 1 year ago

6cc0f1c2

[reland][quant][pt2e] Annotate GRU module (#103358) (#103526)

jerryzh168 committed 1 year ago

0cd155b0

NCCL process group: avoid workEnqueue when capturing cuda graph (#103503)

bottler committed 1 year ago

02548800

Fix freezing tests (#103531)

eellison committed 1 year ago

25b6b95b

Strengthen partially supported invariant of base for chained sources (#103445)

voznesenskym committed 1 year ago

056bf951

Add type hint for retains_grad (#103528)

ringohoffman committed 1 year ago

bc2caa7f

[pt2] add `SymInt` support for `cosine_similarity` (#103400)

nkaretnikov committed 1 year ago

d38b651d

[pt2] add `SymInt` support for `bilinear` (#103396)

nkaretnikov committed 1 year ago

c0763443

[pt2] add metas for `avg_pool3d` and `avg_pool3d_backward` (#103392)

nkaretnikov committed 1 year ago

4a76fb49

[export] Serialize symbolic values (#103273)

angelayi committed 1 year ago

8dc60010

[ONNX] Add constant folding for Softmax op (#102861)

ilyasher committed 1 year ago

876695d4

Always register SHAPE_ENV guard (#103521)

ezyang committed 1 year ago

3804eb10

torch.compiler public namespace (#102182)

msaroufim committed 1 year ago

ea384cd3

Always apply new_empty special case in Dynamo (#103378)

ezyang committed 1 year ago

3596a853

[FSDP][2/n] add use_dtensor flag to both StateDictConfig and OptimStateDictConfig (#103477)

wz337 committed 1 year ago

51d21ffd

Unified aa_filter and aa_filter_075 for bicubic upsampling (#103510)

vfdev-5 committed 1 year ago

72931759

[FSDP] Fix `device_id` when buffer-only module (#103504)

awgu committed 1 year ago

71b56020

Use free_symbols to determine if convolutions involve dynamic shapes (#103486)

ezyang committed 1 year ago

1628bbec

Stop disabling ShapeProp with dynamic_shapes for mkldnn (#103381)

ezyang committed 1 year ago

38890e1d

Detect symbolic tracing_mode with free_symbols (#103515)

ezyang committed 1 year ago

1506aceb

Enable Python dispatcher when ShapeProp with fake mode (#103512)

ezyang committed 1 year ago

ddb682f6

Don't test dynamic_shapes in profiler (#103516)

ezyang committed 1 year ago

af7bd409

Register more foreach op lowerings (#102654)

mlazos committed 1 year ago

05c01b9b

[FSDP] Workaround for GLOO's lack of all_gather_into_tensor. (#103170)

Rodrigo Kumpera committed 1 year ago

5b33d391

Wrong type when exporting {zeros, ones, full, empty, rand, randn}_like ops to onnx (#103048)

tungld committed 1 year ago

b77f1b0f

Fix rerun disabled test uploading logic (#103476)

huydhn committed 1 year ago

e9f2921b

Newer Older