Detoxification for LLM: From Dataset Itself

2026-04-21

AI Summary

This work proposes Hierarchical Semantic-Preserving Detoxification (HSPD), a novel framework that, for the first time, enables end-to-end, semantics-preserving detoxification directly at the pretraining corpus level, addressing the limitation of existing approaches that primarily operate during post-training or inference stages and fail to eliminate toxicity at its source. HSPD leverages Soft Contrastive Decoding (SoCD) to guide large language models in precisely identifying and rewriting toxic segments, producing high-quality detoxified text that can directly replace the original data—thereby circumventing the need for complex downstream alignment procedures. Experimental results demonstrate that HSPD reduces toxicity probability from 0.42 to 0.18 and expected maximum toxicity from 0.43 to 0.20 on GPT2-XL, while achieving state-of-the-art detoxification performance across multiple models, including LLaMA2-7B, OPT-6.7B, and Falcon-7B.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Existing detoxification methods for large language models mainly focus on post-training stage or inference time, while few tackle the source of toxicity, namely, the dataset itself. Such training-based or controllable decoding approaches cannot completely suppress the model's inherent toxicity, whereas detoxifying the pretraining dataset can fundamentally reduce the toxicity that the model learns during training. Hence, we attempt to detoxify directly on raw corpora with SoCD (Soft Contrastive Decoding), which guides an LLM to localize and rewrite toxic spans in raw data while preserving semantics, in our proposed HSPD (Hierarchical Semantic-Preserving Detoxification) pipeline, yielding a detoxified corpus that can drop-in replace the original for fine-tuning or other training. On GPT2-XL, HSPD attains state-of-the-art detoxification, reducing Toxicity Probability (TP) from 0.42 to 0.18 and Expected Maximum Toxicity (EMT) from 0.43 to 0.20. We further validate consistent best-in-class results on LLaMA2-7B, OPT-6.7B, and Falcon-7B. These findings show that semantics-preserving, corpus-level rewriting with HSPD effectively suppresses downstream toxicity while retaining data utility and allowing seamless source-level mitigation, thereby reducing the cost of later model behavior adjustment. (Code is available at: https://github.com/ntsw2001/data_detox_for_llm)

Problem

• detoxification
• large language models
• pretraining dataset
• toxicity
• corpus-level

Innovation

• dataset detoxification
• semantic-preserving rewriting
• Soft Contrastive Decoding
• corpus-level mitigation
• large language models