Lightning-AI / Lightning-AI/pytorch-lightning
FSDP checkpointing uses deprecated APIs with PyTorch 2.2
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 31.4k
- Forks
- 3.8k
- Avg merge
- 6d 7h
- Merged PRs (30d)
- 6
Description
### Bug description
See added deprecation warnings in https://github.com/pytorch/pytorch/pull/113867
### What version are you seeing the problem on?
v2.2
### How to reproduce the bug
Originated from
https://github.com/Lightning-AI/pytorch-lightning/blob/b097a4df3f3fa8b4465861ccab17a44a8ae1ebb9/src/lightning/fabric/strategies/fsdp.py#L496
We already use the newer API for loading
https://github.com/Lightning-AI/pytorch-lightning/blob/b097a4df3f3fa8b4465861ccab17a44a8ae1ebb9/src/lightning/fabric/strategies/fsdp.py#L563-L566
### Error messages and logs
```
/home/carlos/nightly-env/lib/python3.10/site-packages/torch/distributed/checkpoint/state_dict_saver.py:31: UserWarning: 'save_state_dict' is deprecated and will be removed in future versions.Please use 'save' instead.
warnings.warn(
```
### Environment
_No response_
### More info
_No response_
cc @awaelchli @carmocca
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start in src/lightning/fabric/strategies/fsdp.py around line 496 and compare the saving path with the newer loading API around lines 563-566. Update the deprecated checkpointing usage indicated by the warning, then verify FSDP checkpoint saving no longer emits the deprecation warning under PyTorch 2.2.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Activity status
- Stale
- Clarity
- Clearly specified
- Newbie friendliness
- 35/100