Integrative Decoding: Improve Factuality via Implicit Self-consistency

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cheng, Yi, Liang, Xiao, Gong, Yeyun, Xiao, Wen, Wang, Song, Zhang, Yuji, Hou, Wenjun, Xu, Kaishuai, Liu, Wenge, Li, Wenjie, Jiao, Jian, Chen, Qi, Cheng, Peng, Xiong, Wayne
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929684357316608
author Cheng, Yi
Liang, Xiao
Gong, Yeyun
Xiao, Wen
Wang, Song
Zhang, Yuji
Hou, Wenjun
Xu, Kaishuai
Liu, Wenge
Li, Wenjie
Jiao, Jian
Chen, Qi
Cheng, Peng
Xiong, Wayne
author_facet Cheng, Yi
Liang, Xiao
Gong, Yeyun
Xiao, Wen
Wang, Song
Zhang, Yuji
Hou, Wenjun
Xu, Kaishuai
Liu, Wenge
Li, Wenjie
Jiao, Jian
Chen, Qi
Cheng, Peng
Xiong, Wayne
contents Self-consistency-based approaches, which involve repeatedly sampling multiple outputs and selecting the most consistent one as the final response, prove to be remarkably effective in improving the factual accuracy of large language models. Nonetheless, existing methods usually have strict constraints on the task format, largely limiting their applicability. In this paper, we present Integrative Decoding (ID), to unlock the potential of self-consistency in open-ended generation tasks. ID operates by constructing a set of inputs, each prepended with a previously sampled response, and then processes them concurrently, with the next token being selected by aggregating of all their corresponding predictions at each decoding step. In essence, this simple approach implicitly incorporates self-consistency in the decoding objective. Extensive evaluation shows that ID consistently enhances factuality over a wide range of language models, with substantial improvements on the TruthfulQA (+11.2%), Biographies (+15.4%) and LongFact (+8.5%) benchmarks. The performance gains amplify progressively as the number of sampled responses increases, indicating the potential of ID to scale up with repeated sampling.
format Preprint
id arxiv_https___arxiv_org_abs_2410_01556
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Integrative Decoding: Improve Factuality via Implicit Self-consistency
Cheng, Yi
Liang, Xiao
Gong, Yeyun
Xiao, Wen
Wang, Song
Zhang, Yuji
Hou, Wenjun
Xu, Kaishuai
Liu, Wenge
Li, Wenjie
Jiao, Jian
Chen, Qi
Cheng, Peng
Xiong, Wayne
Computation and Language
Artificial Intelligence
Machine Learning
Self-consistency-based approaches, which involve repeatedly sampling multiple outputs and selecting the most consistent one as the final response, prove to be remarkably effective in improving the factual accuracy of large language models. Nonetheless, existing methods usually have strict constraints on the task format, largely limiting their applicability. In this paper, we present Integrative Decoding (ID), to unlock the potential of self-consistency in open-ended generation tasks. ID operates by constructing a set of inputs, each prepended with a previously sampled response, and then processes them concurrently, with the next token being selected by aggregating of all their corresponding predictions at each decoding step. In essence, this simple approach implicitly incorporates self-consistency in the decoding objective. Extensive evaluation shows that ID consistently enhances factuality over a wide range of language models, with substantial improvements on the TruthfulQA (+11.2%), Biographies (+15.4%) and LongFact (+8.5%) benchmarks. The performance gains amplify progressively as the number of sampled responses increases, indicating the potential of ID to scale up with repeated sampling.
title Integrative Decoding: Improve Factuality via Implicit Self-consistency
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2410.01556