pytorch/xla

Pull Requests Commits

Add documentation links to begging of doc for general guidance

pgmoka committed 192 days ago

b12f312a

add comment suggestions

pgmoka committed 192 days ago

803a9297

add symint description

pgmoka committed 193 days ago

9a07371f

add suggestions from comments

pgmoka committed 193 days ago

9c4f1591

Add initial info on autograd config based on knowledge with previous issues

pgmoka committed 194 days ago

c085a214

Add suggestions

pgmoka committed 194 days ago

913804ac

pgmoka committed 195 days ago

dc2f2cb7

Add information about codegen in codegen/xla_native_functions.yaml

pgmoka committed 196 days ago

92fccba3

Fix the ragged kernel bug (#8806)

vanbasten23 committed 197 days ago

Verified 6804c65f

Document xla override (#8802)

pgmoka committed 197 days ago

Verified 0407d6d1

Leverage placeholder tensors for GA API (#8808)

rpsilva-aws committed 197 days ago

Verified 5b7462c6

Manually register einsum on xla (#8801)

pgmoka committed 200 days ago

Verified 1db6c861

Misc fixes (#8804)

qihqi committed 200 days ago

Verified 7aea0585

Re-enable triton tests. (#8803)

ysiraichi committed 201 days ago

Verified 2c70a1c2

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

ysiraichi committed 201 days ago

Verified 6c53a1ef

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

yaochengji committed 201 days ago

Verified d06a9c97

Pin update to 20250303 (#8788)

tengyifei committed 201 days ago

Verified 9b61c1ad

Use placeholder tensor in scan (#8785)

tengyifei committed 202 days ago

Verified a9279287

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

rpsilva-aws committed 202 days ago

Verified 4540d817

Integrate ragged paged attention v2 (#8791)

bythew3i committed 202 days ago

Verified 5644f44a

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

tengyifei committed 203 days ago

Verified 17270e2f

Build torch_xla wheel in build script (#8782)

zpcore committed 203 days ago

Verified 00fac787

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

qihqi committed 204 days ago

Verified cee0820e

Introduce a GRU module implemented with scan (#8777)

tengyifei committed 204 days ago

Verified 0017ff16

Fix build on Python 3.9 (#8759)

tengyifei committed 206 days ago

Verified c76e9497

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

qihqi committed 206 days ago

Verified 60001c74

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

tengyifei committed 206 days ago

Verified 2feb0ac7

Add sm_scale in ragged attention kernel (#8771)

yaochengji committed 207 days ago

Verified 2675e689

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

yaochengji committed 207 days ago

Verified 2e4f073a

Enable default buffer donation for gradient accumulation (#8758)

rpsilva-aws committed 207 days ago

Verified 6712eb91

Older