@inproceedings{huang-etal-2023-large,
title = "Large Language Models Can Self-Improve",
author = "Huang, Jiaxin and
Gu, Shixiang and
Hou, Le and
Wu, Yuexin and
Wang, Xuezhi and
Yu, Hongkun and
Han, Jiawei",
editor = "Bouamor, Houda and
Pino, Juan and
Bali, Kalika",
booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2023",
address = "Singapore",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.emnlp-main.67",
doi = "10.18653/v1/2023.emnlp-main.67",
pages = "1051--1068",
abstract = "Large Language Models (LLMs) have achieved excellent performances in various tasks. However, fine-tuning an LLM requires extensive supervision. Human, on the other hand, may improve their reasoning abilities by self-thinking without external inputs. In this work, we demonstrate that an LLM is also capable of self-improving with only unlabeled datasets. We use a pre-trained LLM to generate {``}high-confidence{''} rationale-augmented answers for unlabeled questions using Chain-of-Though (CoT) prompting and self-consistency, and fine-tune the LLM using those self-generated solutions as target outputs. We show that without any ground truth label, our approach improves the general reasoning ability of a 540B-parameter LLM (74.4{\%}$\rightarrow$82.1{\%} on GSM8K, 90.0{\%}$\rightarrow$94.4{\%} on OpenBookQA, and 63.4{\%}$\rightarrow$67.9{\%} on ANLI-A3) and can also be adapted to extreme low-resource cases where even training questions and CoT prompts are limited. We conduct ablation studies and show that fine-tuning on diverse reasoning paths is critical for self-improvement.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2023-large">
<titleInfo>
<title>Large Language Models Can Self-Improve</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiaxin</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shixiang</namePart>
<namePart type="family">Gu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Le</namePart>
<namePart type="family">Hou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuexin</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuezhi</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongkun</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiawei</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Houda</namePart>
<namePart type="family">Bouamor</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juan</namePart>
<namePart type="family">Pino</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kalika</namePart>
<namePart type="family">Bali</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Singapore</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Large Language Models (LLMs) have achieved excellent performances in various tasks. However, fine-tuning an LLM requires extensive supervision. Human, on the other hand, may improve their reasoning abilities by self-thinking without external inputs. In this work, we demonstrate that an LLM is also capable of self-improving with only unlabeled datasets. We use a pre-trained LLM to generate “high-confidence” rationale-augmented answers for unlabeled questions using Chain-of-Though (CoT) prompting and self-consistency, and fine-tune the LLM using those self-generated solutions as target outputs. We show that without any ground truth label, our approach improves the general reasoning ability of a 540B-parameter LLM (74.4%\rightarrow82.1% on GSM8K, 90.0%\rightarrow94.4% on OpenBookQA, and 63.4%\rightarrow67.9% on ANLI-A3) and can also be adapted to extreme low-resource cases where even training questions and CoT prompts are limited. We conduct ablation studies and show that fine-tuning on diverse reasoning paths is critical for self-improvement.</abstract>
<identifier type="citekey">huang-etal-2023-large</identifier>
<identifier type="doi">10.18653/v1/2023.emnlp-main.67</identifier>
<location>
<url>https://aclanthology.org/2023.emnlp-main.67</url>
</location>
<part>
<date>2023-12</date>
<extent unit="page">
<start>1051</start>
<end>1068</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Large Language Models Can Self-Improve
%A Huang, Jiaxin
%A Gu, Shixiang
%A Hou, Le
%A Wu, Yuexin
%A Wang, Xuezhi
%A Yu, Hongkun
%A Han, Jiawei
%Y Bouamor, Houda
%Y Pino, Juan
%Y Bali, Kalika
%S Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing
%D 2023
%8 December
%I Association for Computational Linguistics
%C Singapore
%F huang-etal-2023-large
%X Large Language Models (LLMs) have achieved excellent performances in various tasks. However, fine-tuning an LLM requires extensive supervision. Human, on the other hand, may improve their reasoning abilities by self-thinking without external inputs. In this work, we demonstrate that an LLM is also capable of self-improving with only unlabeled datasets. We use a pre-trained LLM to generate “high-confidence” rationale-augmented answers for unlabeled questions using Chain-of-Though (CoT) prompting and self-consistency, and fine-tune the LLM using those self-generated solutions as target outputs. We show that without any ground truth label, our approach improves the general reasoning ability of a 540B-parameter LLM (74.4%\rightarrow82.1% on GSM8K, 90.0%\rightarrow94.4% on OpenBookQA, and 63.4%\rightarrow67.9% on ANLI-A3) and can also be adapted to extreme low-resource cases where even training questions and CoT prompts are limited. We conduct ablation studies and show that fine-tuning on diverse reasoning paths is critical for self-improvement.
%R 10.18653/v1/2023.emnlp-main.67
%U https://aclanthology.org/2023.emnlp-main.67
%U https://doi.org/10.18653/v1/2023.emnlp-main.67
%P 1051-1068
Markdown (Informal)
[Large Language Models Can Self-Improve](https://aclanthology.org/2023.emnlp-main.67) (Huang et al., EMNLP 2023)
ACL
- Jiaxin Huang, Shixiang Gu, Le Hou, Yuexin Wu, Xuezhi Wang, Hongkun Yu, and Jiawei Han. 2023. Large Language Models Can Self-Improve. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 1051–1068, Singapore. Association for Computational Linguistics.