Composing Policy Gradients and Prompt Optimization for Language Model Programs

TL;DR

This paper demonstrates that Group Relative Policy Optimization (GRPO) can be effectively combined with prompt optimization to improve language model programs, achieving significant accuracy gains across various tasks.

Contribution

It introduces a multi-module variant of GRPO that works with modular LM programs and shows how it can be combined with prompt optimization for better performance.

Findings

GRPO and multi-module GRPO empirically compose well with prompt optimization.

Combined methods improve accuracy by 11% on average across tasks.

Open-source implementation available at https://dspy.ai.

Abstract

Group Relative Policy Optimization (GRPO) has proven to be an effective tool for post-training language models (LMs). However, AI systems are increasingly expressed as modular programs that mix together multiple LM calls with distinct prompt templates and other tools, and it is not clear how practitioners can best leverage online RL algorithms like GRPO to improve these systems. We begin to address this challenge by investigating whether it is possible to effectively instantiate GRPO for arbitrary multi-prompt programs and whether it can work robustly as an off-the-shelf optimizer for LM programs using the same abstractions and constraints typically involved for prompt optimization. Our main variant of multi-module GRPO constructs groups from module-level invocations, and we also consider trajectory-level grouping as another natural instantiation. We find for the first time that GRPO (and its multi-module counterpart) empirically composes well with automatic prompt optimization, and together they improve accuracy by 11% on average across classification, many-hop search, and privacy-preserving delegation tasks against the post-trained LM - with 5% gains against prompt optimization on its own. We open-source multi-module GRPO in the DSPy library at https://dspy.ai .