ReFT: Reasoning with Reinforced Fine-Tuning

TL;DR

ReFT enhances reasoning in Large Language Models by combining supervised fine-tuning with reinforcement learning, allowing models to learn from multiple reasoning paths and significantly improve performance on math problem-solving tasks.

Contribution

The paper introduces Reinforced Fine-Tuning (ReFT), a novel method that improves reasoning generalization in LLMs by integrating reinforcement learning with supervised fine-tuning without extra data.

Findings

ReFT outperforms standard SFT on GSM8K, MathQA, and SVAMP datasets.

ReFT's performance can be further improved with inference-time strategies.

ReFT demonstrates superior generalization ability without additional training data.

Abstract

One way to enhance the reasoning capability of Large Language Models (LLMs) is to conduct Supervised Fine-Tuning (SFT) using Chain-of-Thought (CoT) annotations. This approach does not show sufficiently strong generalization ability, however, because the training only relies on the given CoT data. In math problem-solving, for example, there is usually only one annotated reasoning path for each question in the training data. Intuitively, it would be better for the algorithm to learn from multiple annotated reasoning paths given a question. To address this issue, we propose a simple yet effective approach called Reinforced Fine-Tuning (ReFT) to enhance the generalizability of learning LLMs for reasoning, with math problem-solving as an example. ReFT first warmups the model with SFT, and then employs on-line reinforcement learning, specifically the PPO algorithm in this paper, to further fine-tune the model, where an abundance of reasoning paths are automatically sampled given the question and the rewards are naturally derived from the ground-truth answers. Extensive experiments on GSM8K, MathQA, and SVAMP datasets show that ReFT significantly outperforms SFT, and the performance can be potentially further boosted by combining inference-time strategies such as majority voting and re-ranking. Note that ReFT obtains the improvement by learning from the same training questions as SFT, without relying on extra or augmented training questions. This indicates a superior generalization ability for ReFT.