SIMPLEMIX: Frustratingly Simple Mixing of Off- and On-policy Data in Language Model Preference Learning.

Tianjian Li, Daniel Khashabi

Browse the full ICML paper archive.