Lightning-AI / Lightning-AI/pytorch-lightning

Revamp Async Checkpointing to Use PyTorch Distributed Checkpoint (DCP) async_save

Open
#21,557 0 comments 0 reactions 1 assignee View on GitHub

@deependujha is already working on this.

Since Feb 27, 2026.

callback: model checkpoint checkpointing feature
Dominant language
Python
Stars
31.4k
Forks
3.8k
Avg merge
6d 7h
Merged PRs (30d)
6

Description

### Description & Motivation

PyTorch Lightning’s current async checkpointing implementation predates PyTorch’s Distributed Checkpoint (DCP) API and feels outdated.

This issue proposes evaluating and migrating Lightning’s async checkpoint logic to leverage `torch.distributed.checkpoint` (DCP), specifically `async_save`, to:

* Align with upstream PyTorch checkpointing APIs
* Improve robustness and maintainability
* Better support distributed and sharded training setups
* Reduce custom logic that duplicates upstream functionality

### Pitch

Use [PyTorch DCP's async_save](https://docs.pytorch.org/docs/stable/distributed.checkpoint.html#torch.distributed.checkpoint.state_dict_saver.async_save)

### Alternatives

_No response_

### Additional context

https://docs.pytorch.org/docs/stable/distributed.checkpoint.html#distributed-checkpoint-torch-distributed-checkpoint

cc @lantiga

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.