Pr eval lora (#515)
* Support evaluation using non-merged peft(lora) weight
* Merge lora into base model to enable fsdp
* Update TUTORIAL.md
* Add setting of `lora_merge` in eval.py
* Remove part of merging lora weight into base
* pinned down peft to avoid accelerate issues
* eval merge fixes
* type hints for load peft
* format string
---------
Co-authored-by: SingL3 <thanks3q@gmail.com>
Co-authored-by: Jeremy D <115047575+bmosaicml@users.noreply.github.com>
