Saved in:
Bibliographic Details
Main Authors: Xu, Wenda, Deutsch, Daniel, Finkelstein, Mara, Juraska, Juraj, Zhang, Biao, Liu, Zhongtao, Wang, William Yang, Li, Lei, Freitag, Markus
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2311.09336
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912084675002368
author Xu, Wenda
Deutsch, Daniel
Finkelstein, Mara
Juraska, Juraj
Zhang, Biao
Liu, Zhongtao
Wang, William Yang
Li, Lei
Freitag, Markus
author_facet Xu, Wenda
Deutsch, Daniel
Finkelstein, Mara
Juraska, Juraj
Zhang, Biao
Liu, Zhongtao
Wang, William Yang
Li, Lei
Freitag, Markus
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2311_09336
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle LLMRefine: Pinpointing and Refining Large Language Models via Fine-Grained Actionable Feedback
Xu, Wenda
Deutsch, Daniel
Finkelstein, Mara
Juraska, Juraj
Zhang, Biao
Liu, Zhongtao
Wang, William Yang
Li, Lei
Freitag, Markus
Computation and Language
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.
title LLMRefine: Pinpointing and Refining Large Language Models via Fine-Grained Actionable Feedback
topic Computation and Language
url https://arxiv.org/abs/2311.09336