ReST meets ReAct: Self-Improvement for Multi-Step Reasoning LLM Agent
This addresses failure cases in complex question-answering systems for AI researchers and practitioners, offering a more efficient and trainable approach.
The paper tackles the problem of multi-step reasoning in LLM agents by introducing a self-improvement method that combines ReAct-style reasoning with ReST-like iterative training, resulting in a fine-tuned small model that achieves comparable performance on challenging benchmarks with two orders of magnitude fewer parameters after just two iterations.
Answering complex natural language questions often necessitates multi-step reasoning and integrating external information. Several systems have combined knowledge retrieval with a large language model (LLM) to answer such questions. These systems, however, suffer from various failure cases, and we cannot directly train them end-to-end to fix such failures, as interaction with external knowledge is non-differentiable. To address these deficiencies, we define a ReAct-style LLM agent with the ability to reason and act upon external knowledge. We further refine the agent through a ReST-like method that iteratively trains on previous trajectories, employing growing-batch reinforcement learning with AI feedback for continuous self-improvement and self-distillation. Starting from a prompted large model and after just two iterations of the algorithm, we can produce a fine-tuned small model that achieves comparable performance on challenging compositional question-answering benchmarks with two orders of magnitude fewer parameters.