LLMRefine: Pinpointing and Refining Large Language Models via Fine-Grained Actionable Feedback
This addresses the challenge of improving LLM generation quality without expensive human feedback, offering a domain-specific solution for text generation tasks.
The paper tackles the problem of costly human feedback for large language models by proposing LLMRefine, an inference-time optimization method that uses a learned fine-grained feedback model to pinpoint defects and iteratively refine outputs, achieving improvements such as up to 1.7 MetricX points on translation tasks and 8.1 ROUGE-L on ASQA.
Recent large language models (LLM) are leveraging human feedback to improve their generation quality. However, human feedback is costly to obtain, especially during inference. In this work, we propose LLMRefine, an inference time optimization method to refine LLM's output. The core idea is to use a learned fine-grained feedback model to pinpoint defects and guide LLM to refine them iteratively. Using original LLM as a proposal of edits, LLMRefine searches for defect-less text via simulated annealing, trading off the exploration and exploitation. We conduct experiments on three text generation tasks, including machine translation, long-form question answering (QA), and topical summarization. LLMRefine consistently outperforms all baseline approaches, achieving improvements up to 1.7 MetricX points on translation tasks, 8.1 ROUGE-L on ASQA, 2.2 ROUGE-L on topical summarization.