Offline RL for Natural Language Generation with Implicit Language Q Learning

Large language models distill broad knowledge from text corpora. However, they can be inconsistent when it comes to completing user specified tasks. This issue can be addressed by finetuning such models via supervised learning on curated datasets, or via reinforcement learning. In this work, we prop...

Full description

Saved in:
Bibliographic Details
Published inarXiv.org
Main Authors Snell, Charlie, Kostrikov, Ilya, Su, Yi, Yang, Mengjiao, Levine, Sergey
Format Paper
LanguageEnglish
Published Ithaca Cornell University Library, arXiv.org 01.05.2023
Subjects
Algorithms
Datasets
Dynamic programming
Dynamic stability
Empirical analysis
Language
Machine learning
Natural language
Optimization
Speech recognition
Online AccessGet full text

Cover

More Information
Summary:Large language models distill broad knowledge from text corpora. However, they can be inconsistent when it comes to completing user specified tasks. This issue can be addressed by finetuning such models via supervised learning on curated datasets, or via reinforcement learning. In this work, we propose a novel offline RL method, implicit language Q-learning (ILQL), designed for use on language models, that combines both the flexible utility maximization framework of RL algorithms with the ability of supervised learning to leverage previously collected data, as well as its simplicity and stability. Our method employs a combination of value conservatism alongside an implicit dataset support constraint in learning value functions, which are then used to guide language model generations towards maximizing user-specified utility functions. In addition to empirically validating ILQL, we present a detailed empirical analysis of situations where offline RL can be useful in natural language generation settings, demonstrating how it can be a more effective utility optimizer than prior approaches for end-to-end dialogue, and how it can effectively optimize high variance reward functions based on subjective judgement, such as whether to label a comment as toxic or not.
ISSN:2331-8422