strisunshinewentingwang/pytorch

Pull Requests Commits

responded to comments

Chillee committed 2 years ago

a20b33bf

Updated to simply interrupt and not resurrect

Chillee committed 2 years ago

f0906baf

try modifying assert?

Chillee committed 2 years ago

6348b365

added more comments

Chillee committed 2 years ago

fd06e184

modified comments

Chillee committed 2 years ago

3f6c8c29

modify check to make it consistent (not sure if this works??)

Chillee committed 2 years ago

fb68b0a5

add some asserts to python_variable and check for resurrections in tp_clear

Chillee committed 2 years ago

dd62378d

fix fetch-depth: 1

clee2000 committed 2 years ago

cdbd39ba

Disallow calling tolist on tensors with nullptr storage

anjali411 committed 2 years ago

ce7feead

[FSDP] Fix `no_sync()` + `FULL_SHARD` root all-gather behavior

awgu committed 2 years ago

3577ff63

[FSDP][Easy] Minor simplifications

awgu committed 2 years ago

fc2778cd

avoid some unnecessary view_copy calls

bdhirsh committed 2 years ago

7ca03dcd

ci: Add credentials to upload test stats for rocm

seemethere committed 2 years ago

e70fea8e

fix unfold for meta tensors

bdhirsh committed 2 years ago

7be1b296

teach ivalue about List[Optional[Tensor]], fix fallbacks

bdhirsh committed 2 years ago

204df13d

fix out= op handling for functionalization

bdhirsh committed 2 years ago

4c7b4b57

split out functionalization codegen to use view_copy operators

bdhirsh committed 2 years ago

cb17973a

fix local_thread:pytorch embeddingbag

jiyuanzFB committed 2 years ago

2602a5e7

[quant][core][gpu][bux fix] Added clone and contiguous() to broadcasted_bias tensor in quantized cudnn linear op

dzdang committed 2 years ago

381e7259

[quant][core][gpu][bug-fix] Added additional caching support in quantized cudnn add_relu op

dzdang committed 2 years ago

6dc71461

Revert D35679120: Add first version of Buck build workflow

malfet committed 2 years ago

c4cf51de

[FSDP] Separate shared code into hooks

rohan-varma committed 2 years ago

9e864f47

[GHF] Fix sync-branches

malfet committed 2 years ago

b582472a

remove fp16 support from cpu linalg functions

ngimel committed 2 years ago

f120d5be

[NVFuser] don't decompose linear if we don't have shape info

davidberard98 committed 2 years ago

ebb60a8b

[FSDP] Add `scatter_full_optim_state_dict()`

awgu committed 2 years ago

019259a6

Revert "Use batched operations for PowerSGD"

pytorchmergebot committed 2 years ago

c5d57e7b

Use batched operations for PowerSGD

MagiaSN committed 2 years ago

5654e633

Convert DDP parameters to ReplicatedTensor during forward pass.

pritamdamania committed 2 years ago

3a38f175

Fixes combinations throwing a meshgrid warning

Mike Ruberry committed 2 years ago

f4d89aa2

Older