Exploring RL-based LLM Training for Formal Language Tasks with Programmed Rewards

TL;DR

This paper explores the use of Proximal Policy Optimization for direct reinforcement learning of large language models on formal language tasks, highlighting challenges and proposing a regularization method to improve exploration.

Contribution

It investigates the feasibility of direct RL from programmed rewards for formal language tasks and introduces a batch-entropy regularization to enhance training stability.

Findings

Pure RL training is challenging for formal language tasks.

A novel batch-entropy regularization aids exploration.

Direct RL may be better for minor adjustments than new tasks.

Abstract

Proximal Policy Optimization (PPO) is commonly used in Reinforcement Learning from Human Feedback to align large language models (LLMs) with downstream tasks. This paper investigates the feasibility of using PPO for direct reinforcement learning (RL) from explicitly programmed reward signals, as opposed to indirect learning from human feedback via an intermediary reward model. We focus on tasks expressed through formal languages, such as mathematics and programming, where explicit reward functions can be programmed to automatically assess the quality of generated outputs. We apply this approach to a sentiment alignment task, a simple arithmetic task, and a more complex game synthesis task. The sentiment alignment task replicates prior research and serves to validate our experimental setup. Our results show that pure RL-based training for the two formal language tasks is challenging, with success being limited even for the simple arithmetic task. We propose a novel batch-entropy regularization term to aid exploration, although training is not yet entirely stable. Our findings suggest that direct RL training of LLMs may be more suitable for relatively minor changes, such as alignment, than for learning new tasks altogether, even if an informative reward signal can be expressed programmatically.