KL-Regularised Q-Learning: A Token-level Action-Value perspective on Online RLHF

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Brown, Jason R, Wells, Lennie, Young, Edward James, Bacallado, Sergio
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866908500068663296
author Brown, Jason R
Wells, Lennie
Young, Edward James
Bacallado, Sergio
author_facet Brown, Jason R
Wells, Lennie
Young, Edward James
Bacallado, Sergio
contents Proximal Policy Optimisation (PPO) is an established and effective policy gradient algorithm used for Language Model Reinforcement Learning from Human Feedback (LM-RLHF). PPO performs well empirically but has a heuristic motivation and handles the KL-divergence constraint used in LM-RLHF in an ad-hoc manner. In this paper, we develop a a new action-value RL method for the LM-RLHF setting, KL-regularised Q-Learning (KLQ). We then show that our method is equivalent to a version of PPO in a certain specific sense, despite its very different motivation. Finally, we benchmark KLQ on two key language generation tasks -- summarisation and single-turn dialogue. We demonstrate that KLQ performs on-par with PPO at optimising the LM-RLHF objective, and achieves a consistently higher win-rate against PPO on LLM-as-a-judge evaluations.
format Preprint
id arxiv_https___arxiv_org_abs_2508_17000
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle KL-Regularised Q-Learning: A Token-level Action-Value perspective on Online RLHF
Brown, Jason R
Wells, Lennie
Young, Edward James
Bacallado, Sergio
Computation and Language
Machine Learning
68T07
I.2.6; I.2.8
Proximal Policy Optimisation (PPO) is an established and effective policy gradient algorithm used for Language Model Reinforcement Learning from Human Feedback (LM-RLHF). PPO performs well empirically but has a heuristic motivation and handles the KL-divergence constraint used in LM-RLHF in an ad-hoc manner. In this paper, we develop a a new action-value RL method for the LM-RLHF setting, KL-regularised Q-Learning (KLQ). We then show that our method is equivalent to a version of PPO in a certain specific sense, despite its very different motivation. Finally, we benchmark KLQ on two key language generation tasks -- summarisation and single-turn dialogue. We demonstrate that KLQ performs on-par with PPO at optimising the LM-RLHF objective, and achieves a consistently higher win-rate against PPO on LLM-as-a-judge evaluations.
title KL-Regularised Q-Learning: A Token-level Action-Value perspective on Online RLHF
topic Computation and Language
Machine Learning
68T07
I.2.6; I.2.8
url https://arxiv.org/abs/2508.17000