pytorch/xla

Pull Requests Commits

[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

Update nightly versions to 2.7 in README (#8490)

tengyifei committed 1 year ago

Verified 3f915ab8

Add build triggers for 2.6-rc1 (#8485)

tengyifei committed 1 year ago

Verified 8e37d509

Update test_train_spmd_linear_model.py to show clear loss value (#8481)

ManfeiBai committed 1 year ago

Verified c05de860

Update libtpu/xla/jax pin (#8480)

lsy323 committed 1 year ago

Verified 7261c794

Make it easy to switch to nightly libtpu and jax (#8468)

tengyifei committed 1 year ago

Verified 7dd2697a

[AdHoc] Test fix: clear metrics and simplify the replicated counter (#8478)

rpsilva-aws committed 1 year ago

Verified 7ef4a120

Update torch_xla wheels to have the suffix +cxx11 instead of .cxx11 (#8472)

bhavya01 committed 1 year ago

Verified aa56abcc

Register offloading ops as torch ops (#8477)

tengyifei committed 1 year ago

Verified e8720ba6

flash_attention: support also cross attention. (#8427)

dudulightricks committed 1 year ago

Verified 5790092f

[1/N] Use std::string_view (#7297)

cyyever committed 1 year ago

Verified 58be354d

Add autocast support for XlaPatchedLinear (#8421)

aws-nm9 committed 1 year ago

Verified 2021c0fe

Newer Older