strisunshinewentingwang/pytorch

Pull Requests Commits

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

[Easy] Improve error message for meta_mm (#97533)

mrshenli committed 1 year ago

a8f7e0b2

[Kineto] Improve Config Options Part 2 - update to new Kineto Submodule (#97556)

aaronenyeshi committed 1 year ago

b32afbbd

disallow invalid value ranges in torch.testing.make_tensor (#96334)

pmeier committed 1 year ago

129e0390

deprecate low==high in torch.testing.make_tensor (#96333)

pmeier committed 1 year ago

47bfb192

fix low and high in torch.testing.make_tensor for integral inputs (#96124)

pmeier committed 1 year ago

76fb9a1c

only apply domain eps for floating and complex types (#97010)

pmeier committed 1 year ago

779cd1f1

honor low and high for torch.bool in torch.testing.make_tensor (#96332)

pmeier committed 1 year ago

9029361f

fix random mask creation in test_maskedtensor (#97017)

pmeier committed 1 year ago

7602aade

QoL improvements for torch.testing.make_tensor (#96125)

pmeier committed 1 year ago

303eb37e

add proper tests for torch.testing.make_tensor (#96331)

pmeier committed 1 year ago

090af4aa

Revert "Sort: Use cub::WarpMergeSort for small sorts (32 < n <= 128) (#96223)"

pytorchmergebot committed 1 year ago

dbe6da79

fix ignored qualifiers errors (#97443)

mikey dagitses committed 1 year ago

85885301

Inductor: fall back bernoulli on cpu (#97002)

blzheng committed 1 year ago

39c81881

[CI] Add missing --cold-start-latency for the dashboard run (#97547)

desertfire committed 1 year ago

2b75955c

Add `is_causal` API for `TransformerDecoder` (#97166)

janEbert committed 1 year ago

95c166cd

Support per channel tensor with unpacking in QNNPACK (#96268)

kwanghoon-meta committed 1 year ago

92605ee7

[DataPipe] Fix missing imports in DataPipe interface file (#97458)

NivekT committed 1 year ago

c5135ff2

Newer Older