pytorch/xla

Pull Requests Commits

fix complication and runtime check

yaochengji committed 203 days ago

e99636c3

yaochengji committed 203 days ago

67068c40

Fix ragged paged attention v2 to resolve the recompilation issue

yaochengji committed 203 days ago

7dcc8788

Integrate ragged paged attention v2 (#8791)

bythew3i committed 203 days ago

Verified 5644f44a

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

tengyifei committed 205 days ago

Verified 17270e2f

Build torch_xla wheel in build script (#8782)

zpcore committed 205 days ago

Verified 00fac787

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

qihqi committed 205 days ago

Verified cee0820e

Introduce a GRU module implemented with scan (#8777)

tengyifei committed 206 days ago

Verified 0017ff16

Fix build on Python 3.9 (#8759)

tengyifei committed 208 days ago

Verified c76e9497

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

qihqi committed 208 days ago

Verified 60001c74

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

tengyifei committed 208 days ago

Verified 2feb0ac7

Add sm_scale in ragged attention kernel (#8771)

yaochengji committed 208 days ago

Verified 2675e689

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

yaochengji committed 209 days ago

Verified 2e4f073a

Enable default buffer donation for gradient accumulation (#8758)

rpsilva-aws committed 209 days ago

Verified 6712eb91

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

vanbasten23 committed 209 days ago

Verified b18a65fa

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

zpcore committed 210 days ago

Verified 1ab8216a

Add start_trace and stop_trace API in profiler (#8743)

lsy323 committed 210 days ago

Verified b4ba17b5

Add --network=host to TPU docker build command (#8735)

bhavya01 committed 210 days ago

Verified 71d0ce65

Increase tolerance in test_scan_xla_patched_linear (#8749)

tengyifei committed 210 days ago

Verified ac430abf

Extend buffer donation aliasing APIs (#8721)

rpsilva-aws committed 210 days ago

Verified 07976424

Introduce apply_xla_patch_to_nn_linear and test that in a scan (#8739)

tengyifei committed 211 days ago

Verified 6f020aad

Add padding ragged paged attention test (#8741)

vanbasten23 committed 211 days ago

Verified 5e1478d6

Improve the ragged kernel benchmarking script. (#8733)

vanbasten23 committed 212 days ago

Verified 79d1e86b

Add autograd function for mark_sharding (#8723)

bhavya01 committed 216 days ago

Verified 1acc987a

Add information to xla.launch (#8724)

pgmoka committed 216 days ago

Verified a08b7f86

Split page indices in the ragged paged attention. (#8688)

vanbasten23 committed 217 days ago

Verified 0c4e82d6

Bump CUDNN version in docker image. (#8718)

ysiraichi committed 218 days ago

Verified 60160233

Bump XLA commit pin. (#8714)

ysiraichi committed 219 days ago

Verified ca248c7e

Support splitting physical axis in SPMD mesh (#8698)

tengyifei committed 222 days ago

Verified 8a712495

update pin (#8677)

lsy323 committed 223 days ago

Verified 6f423d0b

Older