pytorch/xla

Pull Requests Commits

hardcode the branch

JackCaoG committed 2 years ago

830b224b

reenable the torchaudio build

JackCaoG committed 2 years ago

893320c3

Add cloud build runner from tpu-pytorch-releases project to remote builder IAMs (#4945)

mateuszlewko committed 2 years ago

Verified 2870ac2f

Fix `bazel` coverage file path (#4946)

will-cromar committed 2 years ago

Verified 991ef0fe

Bazel fix test (#4944)

stgpetrovic committed 2 years ago

Verified 7d9d43e3

Bazel fixes (#4943)

stgpetrovic committed 2 years ago

Verified 436d1777

Disable BUILD_CPP_TESTS as a default (#4942)

will-cromar committed 2 years ago

Verified 8da54357

Bazel torchxla (#4755)

stgpetrovic committed 2 years ago

Verified ed212d7e

Add BUNDLE_LIBTPU env var for TPU build (#4938)

mateuszlewko committed 2 years ago

Verified 74eff291

Fix test_mp_save (#4936)

JackCaoG committed 2 years ago

Verified 8db3f891

Increase the default XLA_IR_SHAPE_CACHE_SIZE to accomunate the larger model size (#4937)

JackCaoG committed 2 years ago

Verified 9d217b11

Disable cxx abi in the release wheel. (#4900)

vanbasten23 committed 2 years ago

Verified 489a8ff9

Use dynamic shape support in TFRT TPU and CPU (#4921)

will-cromar committed 2 years ago

Verified 21970512

[Distributed] Make xm.all_gather a single graph in Dynamo (#4922)

alanwaketan committed 2 years ago

Verified 486b32e5

[PJRT] Add `tpu.num_available_devices` to get number of XLA TPU devices (#4927)

will-cromar committed 2 years ago

Verified ef95cb41

use test data for test loader (#4924)

vanbasten23 committed 2 years ago

Verified 045aa264

Remove rendezvous from xm.save (#4926)

will-cromar committed 2 years ago

Verified 4f7f3dcd

[Distributed] Test if xm.all_reduce produces a single dynamo graph (#4920)

alanwaketan committed 2 years ago

Verified 08574298

[Distributed] Adopts traceable all_reduce (#4915)

alanwaketan committed 2 years ago

Verified 6c7ca993

Update op lowering guide (#4925)

cowanmeg committed 2 years ago

Verified 3c00d573

[Distributed] Hide token from Python for _xla_all_reduce path (#4913)

alanwaketan committed 2 years ago

Verified 4994597f

[Distributed] Move the cached all reduce token to C++ (#4912)

alanwaketan committed 2 years ago

Verified 44c2fa04

[PJRT] Make `rendezvous` optional in `xm.save` (#4907)

will-cromar committed 2 years ago

Verified 3f8abbdc

Apply sharding to all compatible input tensors (#4919)

jonb377 committed 2 years ago

Verified a537020f

Skip test/test_zero1.py for TPU (#4917)

alanwaketan committed 2 years ago

Verified b4b4d0dc

Faster multinomial implementation (#4904)

cowanmeg committed 2 years ago

Verified b1862ccc

Revert "Apply sharding to all compatible input tensors (#4872)" (#4911)

alanwaketan committed 2 years ago

Verified fa059c47

[Distributed] Remove sea-of-devices mode (#4905)

alanwaketan committed 2 years ago

Verified 397ea4d5

Add a test to verify nonzero correctness. (#4903)

vanbasten23 committed 2 years ago

Verified 8af5207d

Pin setuptools version to fix vision build (#4906)

JackCaoG committed 2 years ago

Verified 1e54c235

Older