Generated Reference Answers for Language Model Alignment

This dataset contains responses generated for the research presented in the paper Learning from Reference Answers: Versatile Language Model Alignment without Binary Human Preference Data.

The paper introduces RefAlign, a versatile REINFORCE-style alignment algorithm that utilizes language generation evaluation metrics, such as BERTScore, between sampled generations and reference answers as surrogate rewards. This approach enables various alignment scenarios, including safety, confidence, and general preference alignment, without relying on binary human preference data or explicit reward models.

Code: https://github.com/mzhaoshuai/RefAlign

This repository contains responses generated by casperhansen/llama-3.3-70b-instruct-awq given the prompts from HuggingFaceH4/ultrafeedback_binarized.

During generation, we use three random seeds (13, 21, 42) to generate three responses and use the model itself to choose the best response.