pytorch/xla

Pull Requests Commits

Set gradient_as_bucket_view=True in test_train_mp_imagenet (#8558)

tengyifei committed 291 days ago

ea399cd3

Cherrypick #8521 into r2.6 (#8563)

tengyifei committed 291 days ago

Verified 31b2b3b7

Introduce HLO graph bindings (#8564)

rpsilva-aws committed 292 days ago

Verified c23cd63d

Metadata agnostic user computation hash (#8557)

rpsilva-aws committed 295 days ago

Verified 8f5d7bd7

[Computation Hash] Introduce deterministic hash for user computations (#8554)

rpsilva-aws committed 295 days ago

Verified 5ce86092

Add xla autocast support, update autocast APIs in grad checkpointing (#8527)

savitha-aws committed 299 days ago

Verified 28b9b0f8

Cherry pick autocast softmax PR (#8511)

avizon-aws committed 303 days ago

Verified 7d95b391

When modifying IR node, make sure to not lose the read_only bit (#8505) (#8508)

mcuiaws committed 316 days ago

Verified 38ed80e8

xm.save() should not set sync_xla_data=True when sync'ing. (#8484) (#8504)

mcuiaws committed 317 days ago

Verified ef857715

Add xm.xla_device_kind() to return XLA device kind string. (#8493) (#8506)

mcuiaws committed 317 days ago

Verified de9a01ec

Compute and hash buffer_donor_indices for step marker (#8467) (#8503)

mcuiaws committed 317 days ago

Verified cb880ca8

Fix a DDP graph capture issue (#8500)

Chengji Yao committed 317 days ago

Verified 32e3c609

More branch cut changes (#8488)

tengyifei committed 324 days ago

Verified cfe54bf3

Update CI to use PyTorch 2.6 branch (#8487)

tengyifei committed 324 days ago

Verified 142e7027

Update libtpu/xla/jax pin (#8480)

lsy323 committed 325 days ago

Verified 7261c794

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

tengyifei committed 326 days ago

Verified 7dd2697a

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

rpsilva-aws committed 326 days ago

Verified 7ef4a120

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

bhavya01 committed 326 days ago

Verified aa56abcc

Register offloading ops as torch ops (#8477)

tengyifei committed 326 days ago

Verified e8720ba6

flash_attention: support also cross attention. (#8427)

dudulightricks committed 327 days ago

Verified 5790092f

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

cyyever committed 327 days ago

Verified 58be354d

Add autocast support for XlaPatchedLinear (#8421)

aws-nm9 committed 327 days ago

Verified 2021c0fe

[LoweringContext] SPMD propagation (#8471)

rpsilva-aws committed 327 days ago

Verified b068cab9

Modify the deprecation plan for XLA_USE_BF16 (#8474)

rpsilva-aws committed 327 days ago

Verified 18128175

Fix #8475: add libtpu-wheels registry when building docker (#8476)

tengyifei committed 327 days ago

Verified 47f737a7

Add autocast support for einsum (#8420)

aws-nm9 committed 327 days ago

Verified 7b5aca68

Remove Jack and Will from code owner (#8473)

JackCaoG committed 327 days ago

Verified 5c062ea3

Support S32/U32 indices for BWD embedding & Neuron implicit downcast (#8462)

rpsilva-aws committed 330 days ago

Verified 00c0e966

[LoweringContext] Support an optimized parameter mapping for SPMD (#8460)

rpsilva-aws committed 330 days ago

Verified 5d11f660

Improve wheel naming logic in ansible (#8465)

tengyifei committed 330 days ago

Verified cff26e5b

Older