- Combined RL/CL (staged learning)
- New transformer model for molecule optimization
- Full integration of all generators with all algorithmic frameworks (RL, TL)
- Reworked scoring component utilizing a plugin mechanism for easy extension
- TOML configuration file format in addition to JSON: note that format is incompatible with release 3.x
- Major code rewrite
- Single repository for all code