Detoxification of Large Language Models through Output-layer Fusion with a Calibration Model

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tian, Yuanhe, Deng, Mingjie, Jin, Guoqing, Song, Yan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918042829586432
author Tian, Yuanhe
Deng, Mingjie
Jin, Guoqing
Song, Yan
author_facet Tian, Yuanhe
Deng, Mingjie
Jin, Guoqing
Song, Yan
contents Existing approaches for Large language model (LLM) detoxification generally rely on training on large-scale non-toxic or human-annotated preference data, designing prompts to instruct the LLM to generate safe content, or modifying the model parameters to remove toxic information, which are computationally expensive, lack robustness, and often compromise LLMs' fluency and contextual understanding. In this paper, we propose a simple yet effective approach for LLM detoxification, which leverages a compact, pre-trained calibration model that guides the detoxification process of a target LLM via a lightweight intervention in its generation pipeline. By learning a detoxified embedding space from non-toxic data, the calibration model effectively steers the LLM away from generating harmful content. This approach only requires a one-time training of the calibration model that is able to be seamlessly applied to multiple LLMs without compromising fluency or contextual understanding. Experiment results on the benchmark dataset demonstrate that our approach reduces toxicity while maintaining reasonable content expression.
format Preprint
id arxiv_https___arxiv_org_abs_2506_01266
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Detoxification of Large Language Models through Output-layer Fusion with a Calibration Model
Tian, Yuanhe
Deng, Mingjie
Jin, Guoqing
Song, Yan
Computation and Language
Artificial Intelligence
Existing approaches for Large language model (LLM) detoxification generally rely on training on large-scale non-toxic or human-annotated preference data, designing prompts to instruct the LLM to generate safe content, or modifying the model parameters to remove toxic information, which are computationally expensive, lack robustness, and often compromise LLMs' fluency and contextual understanding. In this paper, we propose a simple yet effective approach for LLM detoxification, which leverages a compact, pre-trained calibration model that guides the detoxification process of a target LLM via a lightweight intervention in its generation pipeline. By learning a detoxified embedding space from non-toxic data, the calibration model effectively steers the LLM away from generating harmful content. This approach only requires a one-time training of the calibration model that is able to be seamlessly applied to multiple LLMs without compromising fluency or contextual understanding. Experiment results on the benchmark dataset demonstrate that our approach reduces toxicity while maintaining reasonable content expression.
title Detoxification of Large Language Models through Output-layer Fusion with a Calibration Model
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.01266