Skip to content

Add callbacks, async checkpoints, and pipeline - #4246

Open
ak495867 wants to merge 1 commit into
huggingface:mainfrom
ak495867:main
Open

Add callbacks, async checkpoints, and pipeline#4246
ak495867 wants to merge 1 commit into
huggingface:mainfrom
ak495867:main

Conversation

@ak495867

@ak495867 ak495867 commented Sep 10, 2026

Copy link
Copy Markdown

Introduce a lightweight callback system (src/accelerate/callbacks.py) and integrate it into Accelerator (register/trigger helpers and lifecycle hooks). Add asynchronous checkpointing: capture/write split, staging to CPU, and AsyncCheckpointManager (checkpointing.py) to offload disk writes. Add pipeline parallel utilities (src/accelerate/pipeline_parallel.py) with prepare_pipeline and pipeline_forward, and extend ParallelismConfig to support pp_size. Improve optimizer (optimizer.py) to enable fused/foreach defaults where safe and to emit on_optimizer_step callbacks. Misc: wire callbacks into save/load/log/backward flows and minor API/import adjustments.

What does this PR do?

Fixes # (issue)

Before submitting

  • This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
  • Did you read the contributor guideline,
    Pull Request section?
  • Was this discussed/approved via a Github issue or the forum? Please add a link
    to it if that's the case.
  • Did you make sure to update the documentation with your changes? Here are the
    documentation guidelines, and
    here are tips on formatting docstrings.
  • Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

Introduce a lightweight callback system (src/accelerate/callbacks.py) and integrate it into Accelerator (register/trigger helpers and lifecycle hooks). Add asynchronous checkpointing: capture/write split, staging to CPU, and AsyncCheckpointManager (checkpointing.py) to offload disk writes. Add pipeline parallel utilities (src/accelerate/pipeline_parallel.py) with prepare_pipeline and pipeline_forward, and extend ParallelismConfig to support pp_size. Improve optimizer (optimizer.py) to enable fused/foreach defaults where safe and to emit on_optimizer_step callbacks. Misc: wire callbacks into save/load/log/backward flows and minor API/import adjustments.
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