Iterative Reasoning Preference Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pang, Richard Yuanzhe, Yuan, Weizhe, Cho, Kyunghyun, He, He, Sukhbaatar, Sainbayar, Weston, Jason
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917705150365696
author Pang, Richard Yuanzhe
Yuan, Weizhe
Cho, Kyunghyun
He, He
Sukhbaatar, Sainbayar
Weston, Jason
author_facet Pang, Richard Yuanzhe
Yuan, Weizhe
Cho, Kyunghyun
He, He
Sukhbaatar, Sainbayar
Weston, Jason
contents Iterative preference optimization methods have recently been shown to perform well for general instruction tuning tasks, but typically make little improvement on reasoning tasks (Yuan et al., 2024, Chen et al., 2024). In this work we develop an iterative approach that optimizes the preference between competing generated Chain-of-Thought (CoT) candidates by optimizing for winning vs. losing reasoning steps that lead to the correct answer. We train using a modified DPO loss (Rafailov et al., 2023) with an additional negative log-likelihood term, which we find to be crucial. We show reasoning improves across repeated iterations of this scheme. While only relying on examples in the training set, our approach results in increasing accuracy on GSM8K, MATH, and ARC-Challenge for Llama-2-70B-Chat, outperforming other Llama-2-based models not relying on additionally sourced datasets. For example, we see a large improvement from 55.6% to 81.6% on GSM8K and an accuracy of 88.7% with majority voting out of 32 samples.
format Preprint
id arxiv_https___arxiv_org_abs_2404_19733
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Iterative Reasoning Preference Optimization
Pang, Richard Yuanzhe
Yuan, Weizhe
Cho, Kyunghyun
He, He
Sukhbaatar, Sainbayar
Weston, Jason
Computation and Language
Artificial Intelligence
Iterative preference optimization methods have recently been shown to perform well for general instruction tuning tasks, but typically make little improvement on reasoning tasks (Yuan et al., 2024, Chen et al., 2024). In this work we develop an iterative approach that optimizes the preference between competing generated Chain-of-Thought (CoT) candidates by optimizing for winning vs. losing reasoning steps that lead to the correct answer. We train using a modified DPO loss (Rafailov et al., 2023) with an additional negative log-likelihood term, which we find to be crucial. We show reasoning improves across repeated iterations of this scheme. While only relying on examples in the training set, our approach results in increasing accuracy on GSM8K, MATH, and ARC-Challenge for Llama-2-70B-Chat, outperforming other Llama-2-based models not relying on additionally sourced datasets. For example, we see a large improvement from 55.6% to 81.6% on GSM8K and an accuracy of 88.7% with majority voting out of 32 samples.
title Iterative Reasoning Preference Optimization
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2404.19733