Issues

• Relax the strict 1:1 buffer-device validation in ExecuteReplicated to support OpenXLA's 1:N PjRtMemorySpace mapping

  pytorch/xla#9770 · Filter by author VNDJ opened on Jul 8, 2026
  1

• SPMD: spurious retrace every step due to sharding spec mismatch for unsharded tensors

  pytorch/xla#9755 · Filter by author ajakovljevicTT opened on Mar 2, 2026
  2

• Optimization

  pytorch/xla#9754 · Filter by author signore662-beep opened on Mar 1, 2026
  1

• Is torch_xla going to be deprecated soon?

  pytorch/xla#9752 · Filter by author grib0ed0v opened on Feb 17, 2026
  1

• Why does ReplicateShardedData perform an uncached compile+device execute on each sharded output to return to host

  pytorch/xla#9751 · Filter by author jameszianxuTT opened on Feb 4, 2026

• Why might my program hang indefinitely when I use torch_xla.sync?

  pytorch/xla#9750 · Filter by author oluwatimilehin opened on Feb 2, 2026

• [Bug] Calling torch_xla.device() or setting random seed before use_spmd() produces SIGSEGV for unmarked tensors

  pytorch/xla#9735 · Filter by author Dogacel opened on Jan 20, 2026

• reduce memory requirements for FSDP

  pytorch/xla#9734 · Filter by author samuele-bortolato-mag opened on Jan 16, 2026
  1

• Torchax registering jax as PyTorch accelerator

  pytorch/xla#9730 · Filter by author ajakovljevicTT opened on Dec 15, 2025
  1

• Add priority-aware sharding via mark_sharding to guide Shardy propagation

  pytorch/xla#9729 · Filter by author sshonTT opened on Dec 12, 2025

• all_to_all operation generates invalid HLO

  bugdistributed
  pytorch/xla#9727 · Filter by author rajkthakur opened on Dec 5, 2025
  2

• How is GetOutputShardings supposed to work for PJRT Implementers?

  questionruntimestablehlo
  pytorch/xla#9726 · Filter by author jameszianxuTT opened on Dec 2, 2025
  4

• Why isn't there a binding for clearing the XLAGraphExecutor::ComputationCache?

  questionruntime
  pytorch/xla#9712 · Filter by author jameszianxuTT opened on Nov 14, 2025
  3

• torch.inverse: traces successfully, but fails on lowering.

  buglowering
  pytorch/xla#9707 · Filter by author ysiraichi opened on Nov 11, 2025

• torch.distributed.all_reduce works extremely slow with torch_xla

  2.5 releaseperformance
  pytorch/xla#9696 · Filter by author Arioll opened on Oct 29, 2025
  4

• RFC: Evolving PyTorch/XLA for a more native experience on TPU

  RFC
  pytorch/xla#9684 · Filter by author qcc4cp opened on Oct 20, 2025
  21

• Improve PyTorch/XLA Documentation and Clarify SPMD Usage

  distributeddocumentation
  pytorch/xla#9681 · Filter by author Muinez opened on Oct 19, 2025
  1

• PJRT Computation Client Teardown Function

  question
  pytorch/xla#9679 · Filter by author jameszianxuTT opened on Oct 16, 2025
  2

• SIGSEGV crash during multi-host training on TPU VM (v4-32)

  bugdistributedxla:tpu
  pytorch/xla#9677 · Filter by author Locutusque opened on Oct 14, 2025
  5

• shard_as test is failing

  bugtesting
  pytorch/xla#9671 · Filter by author bhavya01 opened on Oct 6, 2025

• all_reduce does not apply scale when xr.world_size == 1

  distributedquestion
  pytorch/xla#9670 · Filter by author afzalxo opened on Oct 6, 2025
  2

• PJRT_Client_Destroy is not invoked in PT/XLA 2.8.0+

  backport_2.9bugruntime
  pytorch/xla#9669 · Filter by author rajkthakur opened on Oct 3, 2025
  11

• Add XLAHooksInterface to PyTorch for XLA

  enhancement
  pytorch/xla#9667 · Filter by author iwknow opened on Oct 2, 2025

• undefined symbol: _ZN3c104cuda10set_deviceEa when trying to use torch_xla 2.8

  2.8 releasebuginstall
  pytorch/xla#9656 · Filter by author jrfonseca opened on Sep 25, 2025
  7

• Correct behavior of torch.ops.xla.write_mlir_debuginfo

  questionstablehlo
  pytorch/xla#9646 · Filter by author tlsdmstn56 opened on Sep 16, 2025
  1