shahbuland
commented
1 year ago - Removed reward model from trainer (now just reward function, passed during construction)
- Moved trainer methods like setup_model and setup_optimizer, etc. to base class
- Removed pipeline from trainer (pipeline now just passed in train method)
- Wrapped promps stat tracker config into DDPO (it is DDPO specific)
- Change default config passed to trainer into drlx config
- Added default values to all subconfigs in DRLX config
- Device for VAE needs to be figured out?
- Fix from_cfg in configs
- fix from_dict in configs
- added a default configs
- moved prompt pipeline into pipeline/init.py
- added pick a pic safe pipeline (sfw cleaning done by @Serena1223 )