Adding filtering for optimizer parameters to only trainable parameters

[bcui19]

trafficstars

What does this PR do?

After FSDP-ing the model, the optimizer should only contain parameters that require gradients (non-frozen parameters). This change is nice for LORA

What issue(s) does this change relate to?

https://mosaicml.atlassian.net/browse/CO-2221

Before submitting

• [ ] Have you read the contributor guidelines?
• [ ] Is this change a documentation change or typo fix? If so, skip the rest of this checklist.
• [ ] Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• [ ] Did you update any related docs and document your change?
• [ ] Did you update any related tests and add any new tests related to your change? (see testing)
• [ ] Did you run the tests locally to make sure they pass?
• [x ] Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[danbider]

looks good! will make fsdp lighter for models with some frozen weights

[mvpatel2000]

Closing as out of date