Saved in:
Bibliographic Details
Main Authors: Cai, Tianchi, Song, Xierui, Jiang, Jiyan, Teng, Fei, Gu, Jinjie, Zhang, Guannan
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2312.02554
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911783485177856
author Cai, Tianchi
Song, Xierui
Jiang, Jiyan
Teng, Fei
Gu, Jinjie
Zhang, Guannan
author_facet Cai, Tianchi
Song, Xierui
Jiang, Jiyan
Teng, Fei
Gu, Jinjie
Zhang, Guannan
contents Aligning language models to human expectations, e.g., being helpful and harmless, has become a pressing challenge for large language models. A typical alignment procedure consists of supervised fine-tuning and preference learning. Most preference learning methods, such as RLHF and DPO, depend on pairwise preference data, which inadequately address scenarios where human feedback is point-wise, leading to potential information loss and suboptimal performance. Addressing this gap, we introduce Point-wise Direct Preference Optimization, a novel preference learning method designed to harness point-wise feedback effectively. Our work also uncovers a novel connection between supervised fine-tuning and point-wise preference learning, culminating in Unified Language Model Alignment, a single-step method that unifies the alignment with human demonstrations and point-wise preferences. Extensive experiments on point-wise preference datasets with binary or continuous labels validate the effectiveness of our methods. Our code and a new dataset with high-quality demonstration samples on harmlessness are released.
format Preprint
id arxiv_https___arxiv_org_abs_2312_02554
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle ULMA: Unified Language Model Alignment with Human Demonstration and Point-wise Preference
Cai, Tianchi
Song, Xierui
Jiang, Jiyan
Teng, Fei
Gu, Jinjie
Zhang, Guannan
Machine Learning
Computation and Language
Aligning language models to human expectations, e.g., being helpful and harmless, has become a pressing challenge for large language models. A typical alignment procedure consists of supervised fine-tuning and preference learning. Most preference learning methods, such as RLHF and DPO, depend on pairwise preference data, which inadequately address scenarios where human feedback is point-wise, leading to potential information loss and suboptimal performance. Addressing this gap, we introduce Point-wise Direct Preference Optimization, a novel preference learning method designed to harness point-wise feedback effectively. Our work also uncovers a novel connection between supervised fine-tuning and point-wise preference learning, culminating in Unified Language Model Alignment, a single-step method that unifies the alignment with human demonstrations and point-wise preferences. Extensive experiments on point-wise preference datasets with binary or continuous labels validate the effectiveness of our methods. Our code and a new dataset with high-quality demonstration samples on harmlessness are released.
title ULMA: Unified Language Model Alignment with Human Demonstration and Point-wise Preference
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2312.02554