pytorch/xla

Pull Requests Commits

Proper test coverage for Neuron CollectivePermute (#8818)

rpsilva-aws committed 237 days ago

Verified acc4c739

Fix jax_env_context (#8819)

bhavya01 committed 237 days ago

Verified 2e04cfb4

Update MarkShardingFunction to be AOTAutograd traceable (#8821)

tengyifei committed 237 days ago

Verified c969276e

[Compilation Cache] Add API to query number of cached computation graphs (#8822)

lsy323 committed 237 days ago

Verified bdf30f66

Refactor Computation of jax into creation and call (#8789)

qihqi committed 238 days ago

Verified 53b3ab81

Remove pages_per_seq padding constrains and add MHA support (#8816)

bythew3i committed 238 days ago

Verified af16c210

Fix the ragged kernel bug (#8806)

vanbasten23 committed 239 days ago

Verified 6804c65f

Document xla override (#8802)

pgmoka committed 240 days ago

Verified 0407d6d1

Leverage placeholder tensors for GA API (#8808)

rpsilva-aws committed 240 days ago

Verified 5b7462c6

Manually register einsum on xla (#8801)

pgmoka committed 243 days ago

Verified 1db6c861

Misc fixes (#8804)

qihqi committed 243 days ago

Verified 7aea0585

Re-enable triton tests. (#8803)

ysiraichi committed 243 days ago

Verified 2c70a1c2

Lower `_conj_copy` and `alias` operation. (#8686)

ysiraichi committed 244 days ago

Verified 6c53a1ef

Fix ragged paged attention v2 to resolve the recompilation issue (#8797)

yaochengji committed 244 days ago

Verified d06a9c97

Pin update to 20250303 (#8788)

tengyifei committed 244 days ago

Verified 9b61c1ad

Use placeholder tensor in scan (#8785)

tengyifei committed 244 days ago

Verified a9279287

Use .backward() with in-place grad mutations for the GA API (#8768)

rpsilva-aws committed 244 days ago

Verified 4540d817

Integrate ragged paged attention v2 (#8791)

bythew3i committed 244 days ago

Verified 5644f44a

Minimal support for calling JAX from PyTorch/XLA (#8781)

tengyifei committed 246 days ago

Verified 17270e2f

Build torch_xla wheel in build script (#8782)

zpcore committed 246 days ago

Verified 00fac787

write _shard_map; refactor flash attention to support 5d inputs. (#8730)

qihqi committed 246 days ago

Verified cee0820e

Introduce a GRU module implemented with scan (#8777)

tengyifei committed 247 days ago

Verified 0017ff16

Fix build on Python 3.9 (#8759)

tengyifei committed 249 days ago

Verified c76e9497

Misc changes to make torchax runnable on GPU. (#8756)

qihqi committed 249 days ago

Verified 60001c74

[scan] Make sure inputs into `fn` are not `device_data` IR nodes (#8769)

tengyifei committed 249 days ago

Verified 2feb0ac7

Add sm_scale in ragged attention kernel (#8771)

yaochengji committed 250 days ago

Verified 2675e689

Revert "Change num_seqs type from int to torch.Tensor" (#8767)

yaochengji committed 250 days ago

Verified 2e4f073a

Enable default buffer donation for gradient accumulation (#8758)

rpsilva-aws committed 250 days ago

Verified 6712eb91

Change num_seqs type from int to torch.Tensor (#8736)

vanbasten23 committed 250 days ago

Verified b18a65fa

Lower `as_strided_copy` use fast path with `slice` (#8734)

zpcore committed 251 days ago

Verified 1ab8216a

Newer Older