Goodtriever: Adaptive Toxicity Mitigation with Retrieval-augmented Models

TL;DR

Goodtriever is a retrieval-augmented method for toxicity mitigation in language models that adapts to evolving language, reduces latency, and improves computational efficiency during toxicity-controlled text generation.

Contribution

It introduces a flexible retrieval-based approach for toxicity mitigation that adapts to language evolution and reduces inference latency.

Findings

Achieves 43% latency reduction during inference

Matches state-of-the-art toxicity mitigation performance

Supports toxicity-controlled text generation

Abstract

Considerable effort has been dedicated to mitigating toxicity, but existing methods often require drastic modifications to model parameters or the use of computationally intensive auxiliary models. Furthermore, previous approaches have often neglected the crucial factor of language's evolving nature over time. In this work, we present a comprehensive perspective on toxicity mitigation that takes into account its changing nature. We introduce Goodtriever, a flexible methodology that matches the current state-of-the-art toxicity mitigation while achieving 43% relative latency reduction during inference and being more computationally efficient. By incorporating a retrieval-based approach at decoding time, Goodtriever enables toxicity-controlled text generation. Our research advocates for an increased focus on adaptable mitigation techniques, which better reflect the data drift models face when deployed in the wild. Code and data are available at https://github.com/for-ai/goodtriever.