feat: only upload lora weight (FLock-io#9)

Fix/trl-dependency-and-training-args (FLock-io#8)

* chore: upgrade trl

* fix: replace trainingargument to trlconfig

Feat/support config file for training args (FLock-io#7)

* feat: add support for training args

* fix: fix wrong dataclass name

* chore: update README

* feat: comment out model in training_args to bypass in training

* fix: filter by model size