Removing Prompt-template Bias in Reinforcement Learning from Human Feedback

ACL ARR 2025 February Submission492 Authors

Abstract:

Reinforcement Learning from Human Feedback (RLHF) has become an essential technique for enhancing pre-trained large language models (LLMs) to generate responses that align with human preferences and societal values.
Although RLHF has shown promise, the training of reward models (RMs) still faces the challenge of \emph{reward hacking}, motivating recent works to prevent RMs from finding shortcuts that bypass the intended optimization objectives by identifying simplistic patterns such as response length.
Besides the issue of \emph{length bias}, our work firstly reveals that \emph{prompt-template bias} learned by RMs can also cause \emph{reward hacking} when dealing with some marginal samples, resulting in LLMs preferring to generate responses in a specific format after RLHF fine-tuning, regardless of the format requested in the prompt.
To this end, we propose a low-cost but effective method, namely Prompt Bias Calibration (PBC), to estimate the \emph{prompt-template bias} term during reward modeling, which can be utilized to calibrate reward scores in the following RL fine-tuning process.
Then, we show that our PBC method can be flexibly combined with existing algorithms of removing \emph{length bias}, leading to a further improvement in the aspect of enhancing the quality of generated responses.

Paper Type: Long

Research Area: Language Modeling

Research Area Keywords: LLM, RLHF, reward hacking

Contribution Types: Model analysis & interpretability

Languages Studied: English

Submission Number: 492