[Train] Support FSDP Strategy for LightningTrainer

[woshiyyya]

Why are these changes needed?

Previously, LightningTrainer only support DDP strategy for distributed training. This PR adds support for FSDP strategy.

Related issue number

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
  • I've added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[justinvyu]

Thanks! Left some questions.

[amogkam]

Any plans to add an example, like porting over this GPT fine tuning with PTL+fsdp example https://github.com/SeanNaren/minGPT/tree/fairscale and showing how to do this on multiple nodes?

This would be a really great flagship example for this integration, and aligns well with our LLM initiative.

[woshiyyya]

Any plans to add an example, like porting over this GPT fine tuning with PTL+fsdp example https://github.com/SeanNaren/minGPT/tree/fairscale and showing how to do this on multiple nodes?

This would be a really great flagship example for this integration, and aligns well with our LLM initiative.

Good idea. What do you think @matthewdeng ?

[matthewdeng]

Yeah, we should add an example for 2.5 😃