TRPrompt: Bootstrapping Query-Aware Prompt Optimization from Textual Rewards

TL;DR

TRPrompt introduces a novel framework that leverages textual feedback to train prompt models, enhancing query-specific reasoning in large language models without prior datasets, achieving state-of-the-art results on math datasets.

Contribution

It unifies textual and reward-based prompt optimization methods by directly training prompt models with textual feedback, eliminating the need for pre-collected datasets.

Findings

Achieves state-of-the-art prompts for GSMHard and MATH datasets.

Effectively incorporates textual rewards into prompt training.

Improves reasoning abilities of LLMs without parameter updates.

Abstract

Prompt optimization improves the reasoning abilities of large language models (LLMs) without requiring parameter updates to the target model. Following heuristic-based "Think step by step" approaches, the field has evolved in two main directions: while one group of methods uses textual feedback to elicit improved prompts from general-purpose LLMs in a training-free way, a concurrent line of research relies on numerical rewards to train a special prompt model, tailored for providing optimal prompts to the target model. In this paper, we introduce the Textual Reward Prompt framework (TRPrompt), which unifies these approaches by directly incorporating textual feedback into training of the prompt model. Our framework does not require prior dataset collection and is being iteratively improved with the feedback on the generated prompts. When coupled with the capacity of an LLM to internalize the notion of what a "good" prompt is, the high-resolution signal provided by the textual rewards allows us to train a prompt model yielding state-of-the-art query-specific prompts for the problems from the challenging math datasets GSMHard and MATH.