All pull requests

Pull requests

• test: skip test_noncontiguous_all_to_all_error for backends that accept strided all_to_all input

  #164·aws-charansg opened Nov 11, 2026·

  ·

• Move distributed marker helpers out of conftest

  #158·mgray07 opened Nov 2, 2026·

  ·
  1

• Restore supported Triton guards in Dynamo tests

  #157·mgray07 opened Nov 1, 2026·

  ·

• profiler: recover callbacks after optional integral capture errors

  #156·mgray07 opened Nov 1, 2026·

  ·

• Limit foreach graph tests to supported backends

  #155·mgray07 opened Nov 1, 2026·

  ·

• Order eager references in Dynamo stream tests

  #154·mgray07 opened Nov 1, 2026·

  ·

• Preserve CPU coverage in sparse PrivateUse1 module tests

  #153·mgray07 opened Nov 1, 2026·

  ·

• Use an in-memory store in fake subgroup tests

  #152·mgray07 opened Oct 31, 2026·

  ·

• Scope nearest upsampling launch-limit expectation to CUDA

  #151·mgray07 opened Oct 31, 2026·

  ·

• Scope optimizer peak-memory tests to the CUDA allocator

  #150·mgray07 opened Oct 31, 2026·

  ·

• Prepare wheel-aligned Neuron tests for PyTorch 45f9fe1c

  #149·mgray07 opened Oct 31, 2026·

  ·

• Count only written TorchCI attempts

  #148·mgray07 opened Oct 31, 2026·

  ·

• Restore functorch dispatch state when tensor formatting raises

  #147·mgray07 opened Oct 31, 2026·

  ·

• Preserve quantization metadata through DDP hook futures

  #146·mgray07 opened Oct 31, 2026·

  ·

• Use the registered device backend in ZeRO optimizer tests

  #145·mgray07 opened Oct 31, 2026·

  ·

• Fix PrivateUse1 test helper and device properties

  #144·mgray07 opened Oct 31, 2026·

  ·

• Fix missing TEST_PRIVATEUSE1 import in common_fsdp

  #143·mgray07 opened Oct 31, 2026·

  ·

• test: run TestTraceableCollectives on accelerators with one process per rank (pt_214)

  #142·AbigailFernandes opened Oct 30, 2026Collaborator·

  ·

• test: run test_functional_api accelerator tests on any accelerator, one process per rank

  #141·AbigailFernandes opened Oct 30, 2026Collaborator·

  ·

• test: select the MX scale swizzle from the test device

  #138·ahoffman-aws opened Oct 28, 2026·

  ·
  5

• test(dynamo): device-generic inputs/expectations in hooks and reorder_logs (pt_214)

  #137·aws-alexk opened Oct 15, 2026·

  ·

• test: resolve the PrivateUse1 primary device without running setUpClass at import (pt_214)

  #136·aws-alexk opened Oct 14, 2026·

  ·

• test(dynamo): create inputs on the current accelerator in device-mismatch tests (pt_214)

  #135·aws-alexk opened Oct 14, 2026·

  ·

• Add aten.argwhere.default to meta_dispatch_expected_failures

  #117·aws-zhenguo opened Sep 9, 2026·

  ·
  1

• feat: add aten ops suite for dynamo, wrapped in torch.compile

  #103·andrewd-aws opened Aug 21, 2026·

  ·