GReaTer: Gradients over Reasoning Makes Smaller Language Models Strong Prompt Optimizers

TL;DR

GReaTer introduces a gradient-based prompt optimization method that enhances small language models' reasoning capabilities, outperforming existing techniques and reducing reliance on large, costly LLMs.

Contribution

The paper presents GReaTer, a novel approach that leverages gradient information over reasoning to optimize prompts for lightweight models, bridging the performance gap with larger models.

Findings

GReaTer outperforms previous prompt optimization methods across multiple reasoning tasks.

Optimized prompts by GReaTer achieve transferability and sometimes surpass larger models.

Gradient-based prompt optimization reduces dependence on expensive large language models.

Abstract

The effectiveness of large language models (LLMs) is closely tied to the design of prompts, making prompt optimization essential for enhancing their performance across a wide range of tasks. Many existing approaches to automating prompt engineering rely exclusively on textual feedback, refining prompts based solely on inference errors identified by large, computationally expensive LLMs. Unfortunately, smaller models struggle to generate high-quality feedback, resulting in complete dependence on large LLM judgment. Moreover, these methods fail to leverage more direct and finer-grained information, such as gradients, due to operating purely in text space. To this end, we introduce GReaTer, a novel prompt optimization technique that directly incorporates gradient information over task-specific reasoning. By utilizing task loss gradients, GReaTer enables self-optimization of prompts for open-source, lightweight language models without the need for costly closed-source LLMs. This allows high-performance prompt optimization without dependence on massive LLMs, closing the gap between smaller models and the sophisticated reasoning often needed for prompt refinement. Extensive evaluations across diverse reasoning tasks including BBH, GSM8k, and FOLIO demonstrate that GReaTer consistently outperforms previous state-of-the-art prompt optimization methods, even those reliant on powerful LLMs. Additionally, GReaTer-optimized prompts frequently exhibit better transferability and, in some cases, boost task performance to levels comparable to or surpassing those achieved by larger language models, highlighting the effectiveness of prompt optimization guided by gradients over reasoning. Code of GReaTer is available at https://github.com/psunlpgroup/GreaTer.