Preference Optimization for Reasoning with Pseudo Feedback

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jiao, Fangkai, Guo, Geyang, Zhang, Xingxing, Chen, Nancy F., Joty, Shafiq, Wei, Furu
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917922182529024
author Jiao, Fangkai
Guo, Geyang
Zhang, Xingxing
Chen, Nancy F.
Joty, Shafiq
Wei, Furu
author_facet Jiao, Fangkai
Guo, Geyang
Zhang, Xingxing
Chen, Nancy F.
Joty, Shafiq
Wei, Furu
contents Preference optimization techniques, such as Direct Preference Optimization (DPO), are frequently employed to enhance the reasoning capabilities of large language models (LLMs) in domains like mathematical reasoning and coding, typically following supervised fine-tuning. These methods rely on high-quality labels for reasoning tasks to generate preference pairs; however, the availability of reasoning datasets with human-verified labels is limited. In this study, we introduce a novel approach to generate pseudo feedback for reasoning tasks by framing the labeling of solutions to reason problems as an evaluation against associated test cases. We explore two forms of pseudo feedback based on test cases: one generated by frontier LLMs and the other by extending self-consistency to multi-test-case. We conduct experiments on both mathematical reasoning and coding tasks using pseudo feedback for preference optimization, and observe improvements across both tasks. Specifically, using Mathstral-7B as our base model, we improve MATH results from 58.3 to 68.6, surpassing both NuminaMath-72B and GPT-4-Turbo-1106-preview. In GSM8K and College Math, our scores increase from 85.6 to 90.3 and from 34.3 to 42.3, respectively. Building on Deepseek-coder-7B-v1.5, we achieve a score of 24.6 on LiveCodeBench (from 21.1), surpassing Claude-3-Haiku.
format Preprint
id arxiv_https___arxiv_org_abs_2411_16345
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Preference Optimization for Reasoning with Pseudo Feedback
Jiao, Fangkai
Guo, Geyang
Zhang, Xingxing
Chen, Nancy F.
Joty, Shafiq
Wei, Furu
Computation and Language
Preference optimization techniques, such as Direct Preference Optimization (DPO), are frequently employed to enhance the reasoning capabilities of large language models (LLMs) in domains like mathematical reasoning and coding, typically following supervised fine-tuning. These methods rely on high-quality labels for reasoning tasks to generate preference pairs; however, the availability of reasoning datasets with human-verified labels is limited. In this study, we introduce a novel approach to generate pseudo feedback for reasoning tasks by framing the labeling of solutions to reason problems as an evaluation against associated test cases. We explore two forms of pseudo feedback based on test cases: one generated by frontier LLMs and the other by extending self-consistency to multi-test-case. We conduct experiments on both mathematical reasoning and coding tasks using pseudo feedback for preference optimization, and observe improvements across both tasks. Specifically, using Mathstral-7B as our base model, we improve MATH results from 58.3 to 68.6, surpassing both NuminaMath-72B and GPT-4-Turbo-1106-preview. In GSM8K and College Math, our scores increase from 85.6 to 90.3 and from 34.3 to 42.3, respectively. Building on Deepseek-coder-7B-v1.5, we achieve a score of 24.6 on LiveCodeBench (from 21.1), surpassing Claude-3-Haiku.
title Preference Optimization for Reasoning with Pseudo Feedback
topic Computation and Language
url https://arxiv.org/abs/2411.16345