strisunshinewentingwang/pytorch

Pull Requests Commits

Dynamo size dim kwargs (#97450)

nima10khodaveisi committed 1 year ago

a6bc1f3a

[cpp_extension.py] fix bogus `_check_cuda_version` (#97602)

stas00 committed 1 year ago

8275e5d2

do not insert to_dtype for memory copy only buffers (#97147)

zhuhaozhe committed 1 year ago

a1ada050

Add support for copysign operator in functorch (#96018)

lithomas1 committed 1 year ago

e1f153f3

Remove unnecessary retain_grad call from gradcheck (#96923)

soulitzer committed 1 year ago

d0abc314

Modify all calls to checkpoint pass use_reentrant explicitly (#97376)

soulitzer committed 1 year ago

51c3fd39

Split rnn primitive for inference and training (#96736)

yanbing-j committed 1 year ago

38da54e9

[Dynamo] Unspec int list if enabling dynamic_shapes (#97557)

yanboliang committed 1 year ago

e3df6a7c

Specify file encoding in test_torch.py (#97628)

malfet committed 1 year ago

542fb0b1

[BE] Use nested namespaces in sparse (#97581)

malfet committed 1 year ago

b73e8cd4

add -std=c++17 to windows cuda compilations (#97515)

mikey dagitses committed 1 year ago

461f088c

[BE] Apply ufmt to run_test and GitHub Python util scripts (#97588)

huydhn committed 1 year ago

4c0dce50

[inductor] Fix broadcast of random seed in mm epilogue (#97591)

ngimel committed 1 year ago

f09347a9

Fixes double printing of code in debug mode (#97608)

ngimel committed 1 year ago

4f2ac8ab

[inductor] support SymPy exprs in `reflection_pad2d_backward` lowering (#97604)

nkaretnikov committed 1 year ago

dc45ad70

[inductor] support non-tensor ops with dynamic shapes (#97519)

nkaretnikov committed 1 year ago

9585a7ff

Dynamo stride dim kwargs (#97444)

nima10khodaveisi committed 1 year ago

13dcf635

Add accuracy tests for traced optimizers (#97577)

mrshenli committed 1 year ago

233742cb

Default to aot_eager for torch.compile on MPS (#96980)

msaroufim committed 1 year ago

1b08a013

Enable full train_step tracing and customizable dist graph expansion (#97416)

mrshenli committed 1 year ago

75fb0b6c

Enable lowering to inductor (#96927)

leslie-fang-intel committed 1 year ago

e67b5810

[FSDP] Fix bug in determining whether parameters need to be materialized (#97488)

jamesr66a committed 1 year ago

3b1b585a

[BE] Make `USE_FLASH_ATTENTION` private (#97579)

malfet committed 1 year ago

14177f0d

[vmap] ldl_factor: batch rule (#97518)

kshitij12345 committed 1 year ago

5e014bfb

Rewrite NCCL watchdog to more reliably throw timeout (#97066)

kwen2501 committed 1 year ago

f89af601

Use unordered NEQ comparison for vec512 operator!= implementations (#97466)

bjhargrave committed 1 year ago

ee934fd6

[Better Transformer] make is_causal a hint and force attn_mask to be set on `is_causal=True` in F.MHA (#97214)

mikekgfb committed 1 year ago

c757647d

Add myself to nn codeowners (#97277)

mikaylagawarecki committed 1 year ago

2e8086b0

[NCCL] Cleanup NCCL-no-record streams, move to `TORCH_NCCL_AVOID_RECORD_STREAMS` (#97053)

eqy committed 1 year ago

0781188e

[Easy] Apply black to format _spmd files (#97534)

mrshenli committed 1 year ago

021de486

Newer Older