Work

Prompt Compression and Contrastive Conditioning for Controllability and Toxicity Reduction in Language Models

Year: 2022

Type: conference-paper

Abstract: We explore the idea of compressing the prompts used to condition language models, and show that compressed prompts can retain a substantive amount of information about the original prompt.For severely compressed prompts, while fine-grained information is lost, abstract information and general sentiments can be retained with surprisingly few parameters, which can be useful in the context of decodetime algorithms for controllability and toxicity reduction.We explore contrastive conditioning to steer language model generation towards desirable text and away from undesirable text, and find that some complex prompts can be effectively compressed into a single token to guide generation.We also show that compressed prompts are largely compositional, and can be constructed such that they can be used to control independent aspects of generated text.

Source: Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)

DOI: 10.18653/v1/2022.findings-emnlp.412

Authors: David Wingate, Mohammad Shoeybi, Taylor Sorensen

Institutions: Brigham Young University, University of Washington, Nvidia (United States)

Language: English

Topic: Topic Modeling

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Funders: National Science Foundation, Nvidia

Award: EAGER: Harnessing Accurate Bias in Large-Scale Language Models