Skip to content

test(checkpoint): respect effective pinning capability - #86

Draft
aws-mengchiy wants to merge 1 commit into
release/pt_213from
fix/checkpoint-pin-expectation-pt213
Draft

test(checkpoint): respect effective pinning capability#86
aws-mengchiy wants to merge 1 commit into
release/pt_213from
fix/checkpoint-pin-expectation-pt213

Conversation

@aws-mengchiy

Copy link
Copy Markdown

Summary

Literal cherry-pick of 3dbc406f015 onto release/pt_213.

StateDictStager disables requested pinning when CUDA is unavailable, but the device-generalized test asserted the requested pin_memory flag. Assert against stager.pin_memory, the effective capability-gated value, while retaining all four pin/share configurations and shared-memory coverage.

Validation

  • python -m py_compile test/distributed/checkpoint/test_state_dict_stager.py
  • Trainium with PyTorch 2.12.1+cpu: the exact test passes with --runxfail (1 passed, 4 subtests passed).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant