strisunshinewentingwang/pytorch

Pull Requests Commits

ns for fx: make return type of ns APIs future proof (#52789)

vkuzo committed 3 years ago

d2e88246

ns for fx: unify return types of weight and activation APIs (#52779)

vkuzo committed 3 years ago

fe068157

ns for fx: decouple subgraph names from node names (#52771)

vkuzo committed 3 years ago

7094d970

[caffe2] update load_save_test.py to also verify the chunking behavior

simpkins committed 3 years ago

cd9ac54e

[caffe2] use AddNAlreadyReserved() when serializing blobs

simpkins committed 3 years ago

b4a8d982

[caffe2] move the SaveOp implementation from a header to a .cc file

simpkins committed 3 years ago

3969391c

Update to replace AT_ERROR with TORCH_CHECK (#52711)

lware1m1s committed 3 years ago

fdd25f82

Make NumPy dependency dynamic (#52794)

malfet committed 3 years ago

a0a1bb07

Adding functional way of stacking DataPipes with fixed mypy (#52885)

VitalyFedyunin committed 3 years ago

9a03e654

[FX][EZ] Hoist custom class .so loading into setUp (#52883)

James Reed committed 3 years ago

f40c9db6

[quant] Fix conv packed param serialization in state_dict (#52787)

jerryzh168 committed 3 years ago

6514a473

quant norm layers: move scale + zp to buffers (#52861)

vkuzo committed 3 years ago

a27aaa49

[FX] Use precompiled regex in graph name processing (#52853)

James Reed committed 3 years ago

51d8543a

.github: Add workflow to build conda packages (#51243)

seemethere committed 3 years ago

569d4fe3

`maybe_resize_storage_cuda` new_size argument should be unsigned (#52672)

malfet committed 3 years ago

649760e5

Add sample validation for LKJCholesky.log_prob (#52763)

neerajprad committed 3 years ago

0f3a3f22

Improve test_reference_numerics (#51604)

zasdfgbnm committed 3 years ago

a52001f9

Fix resource leak bug in TCPStore constructor (#52860)

cbalioglu committed 3 years ago

94da8b98

[nnc] Test for direct usage of ramp/broadcast

bertmaher committed 3 years ago

8ba7c491

Fix incorrect runtime error in mul_() when the tensor layout is Mkldnn (#51758)

kazutoshi-ak committed 3 years ago

0b939740

Revert D26644079: [pytorch][PR] Adding functional way of stacking DataPipes

H-Huang committed 3 years ago

da732c76

[vulkan] Add nonVarTypeModeGuard to vulkan tests and speed_benchmark_torch (#52535)

SS-JIA committed 3 years ago

c2558b4b

Use touch() in pathlib for better compatibility on Windows (#52729)

skyline75489 committed 3 years ago

e94940b1

quant: fix conv transpose with qconfig == None (#52844)

vkuzo committed 3 years ago

19a8ada8

Added torch.no_grad() to update_bn (#52654)

MiHarsh committed 3 years ago

c871abec

Upgrade onednn to v.1.8.1 (#51184)

XiaobingSuper committed 3 years ago

0e86f14e

Adding functional way of stacking DataPipes (#52507)

VitalyFedyunin committed 3 years ago

7972036b

Expose Store's timeout and TCPStore's host and port in Python API (#52784)

cbalioglu committed 3 years ago

a11b6011

Test for distributed RL with RPC (#52393)

jbschlosser committed 3 years ago

f974cf46

Fix TensorPipe agent trying to double-set error (#52837)

lw committed 3 years ago

163a91be

Newer Older