How to Evaluate Reward Models for RLHF

TL;DR

This paper introduces PPE, a benchmark for evaluating reward models in RLHF by using proxy tasks to predict downstream language model performance, reducing the need for costly full RLHF training.

Contribution

The paper presents PPE, the first reward model benchmark explicitly linked to real-world human preference outcomes, and develops proxy tasks to efficiently evaluate reward models.

Findings

Proxy tasks correlate with downstream RLHF performance.

PPE provides a practical benchmark for reward model evaluation.

Open-source code enables community use and development.

Abstract

We introduce a new benchmark for reward models that quantifies their ability to produce strong language models through RLHF (Reinforcement Learning from Human Feedback). The gold-standard approach is to run a full RLHF training pipeline and directly probe downstream LLM performance. However, this process is prohibitively expensive. To address this, we build a predictive model of downstream LLM performance by evaluating the reward model on proxy tasks. These proxy tasks consist of a large-scale human preference and a verifiable correctness preference dataset, in which we measure 12 metrics across 12 domains. To investigate which reward model metrics are most correlated to gold-standard RLHF outcomes, we launch an end-to-end RLHF experiment on a large-scale crowdsourced human preference platform to view real reward model downstream performance as ground truth. Ultimately, we compile our data and findings into Preference Proxy Evaluations (PPE), the first reward model benchmark explicitly linked to post-RLHF real-world human preference performance, which we open-source for public use and further development. Our code and evaluations can be found at https://github.com/lmarena/PPE .