Fetching the paper…

Rewarded soups: towards Pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards · Around