pytorch/xla

Pull Requests Commits

[Computation Hash] Introduce deterministic hash for user computations (#8539)

rpsilva-aws committed 1 year ago

Verified 196cab3f

Revert "fix batch_norm amp autocast" (#8547)

tengyifei committed 1 year ago

Verified 05ca2241

Fix typos in pytorch on xla docs (#8543)

MichalPitr committed 1 year ago

Verified de756865

fix batch_norm amp autocast (#8498)

yaochengji committed 1 year ago

Verified 1cdb1eff

doc change: stablehlo.md (#8544)

qihqi committed 1 year ago

Verified 004845f5

[torch_xla2] Update test_ops.py (#8403)

ManfeiBai committed 1 year ago

Verified df02cacf

fix ab in flash attention (#8540)

ZhiyuLi-goog committed 1 year ago

Verified c673809a

Fully revamp aten embedding bag for JAX backend (#8535)

vyom1611 committed 1 year ago

Verified 5278e7a5

pin update (#8536)

lsy323 committed 1 year ago

Verified d36404c2

Use the same decomp table for eager and export (#8531)

qihqi committed 1 year ago

Verified 0c64c365

[scan] Support different output dtypes (#8530)

tengyifei committed 1 year ago

Verified d3d00ac8

[scan] Handle fn output aliasing an input (#8529)

tengyifei committed 1 year ago

Verified a2d3d4d0

Add back the implementation of adaptive_pool_2d (#8526)

qihqi committed 1 year ago

Verified 0ca733b6

Fix DDP with gradient_as_bucket_view enabled (#8521)

Chengji Yao committed 1 year ago

Verified bc3bf1f7

[scan] use device_parameter_id_tensor_mapping (#8524)

tengyifei committed 1 year ago

Verified 0289022d

Add xla autocast support, update autocast APIs in checkpointing (#8523)

savitha-aws committed 1 year ago

Verified 31919d54

[torchxla2] Drop Python 3.9 support (#8494)

chunnienc committed 1 year ago

Verified 40efdb7b

Reenable OpInfo tests. (#8519)

qihqi committed 1 year ago

Verified c3853594

Relax atol of cos; it's failing CI. (#8518)

qihqi committed 1 year ago

Verified cb44d629

Prepare to make 2.6.0 rc3 wheel (#8513)

tengyifei committed 1 year ago

Verified d4c40dc3

[torch_xla2] Fix interpolate and adaptive avg pool (#8512)

dvhg committed 1 year ago

Verified 58bb991c

[torch_xla2] Add transposed conv ops (#8426)

dvhg committed 1 year ago

Verified 5711d1d8

Added softmax to autocast policy (#8509)

avizon-aws committed 1 year ago

Verified 38d08689

When modifying IR node, make sure to not lose the read_only bit (#8505)

mcuiaws committed 1 year ago

Verified e4a9e125

Remove SPMD logging to indicate that we're using virtual device (#8501)

bhavya01 committed 1 year ago

Verified 6ca816d7

xm.save() should not set sync_xla_data=True when sync'ing. (#8484)

mcuiaws committed 1 year ago

Verified d3ed9827

Fix a DDP graph capture issue (#8489)

Chengji Yao committed 1 year ago

Verified b1869a8a

Add xm.xla_device_kind() to return XLA device kind string. (#8493)

mcuiaws committed 1 year ago

Verified 0121444c

Compute and hash buffer_donor_indices for step marker (#8467)

mcuiaws committed 1 year ago

Verified 01548502

Migrate from capture_pre_autograd_graph to export_for_training (#8398)

yushangdi committed 1 year ago

Verified b2b890e9

Newer Older