DSCD: Large Language Model Detoxification with Self-Constrained Decoding
Not on the map. Why?
Abstract (arXiv)
Detoxification in large language models (LLMs) remains a significant research challenge. Existing decoding detoxification methods are all based on external constraints, which require additional resource overhead and lose generation fluency. This work proposes Detoxification with Self-Constrained Decoding (DSCD), a novel method for LLM detoxification without parameter fine-tuning. DSCD strengthens the inner next-token distribution of the safety layer while weakening that of hallucination and toxic layers during output generation. This effectively diminishes toxicity and enhances output safety. DSCD offers lightweight, high compatibility, and plug-and-play capabilities, readily integrating with existing detoxification methods for further performance improvement. Extensive experiments on representative open-source LLMs and public datasets validate DSCD's effectiveness, demonstrating state-of-the-art (SOTA) performance in both detoxification and generation fluency, with superior efficiency compared to existing methods. These results highlight DSCD's potential as a practical and scalable solution for safer LLM deployments.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 28 of its 37 references that are in the map. First posted on arXiv 15 Oct 2025.
Builds on
28 of 37 references.
References · 28
- 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
- 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
- 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,919
- 2022Locating and Editing Factual Associations in GPTcited by 2,609
- 2022Red Teaming Language Models with Language Modelscited by 983
- 2022Mass-Editing Memory in a Transformercited by 976
- 2024Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluatorscited by 865
- 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 828
- 2022Contrastive Decoding: Open-ended Text Generation as Optimizationcited by 598
- 2017BranchyNet: Fast Inference via Early Exiting from Deep Neural Networkscited by 591
- 2024How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMscited by 533
- 2022Memory-Based Model Editing at Scalecited by 443
16 more
- 2023SmoothLLM: Defending Large Language Models Against Jailbreaking Attackscited by 434
- 2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrationscited by 401
- 2023Editing Large Language Models: Problems, Methods, and Opportunitiescited by 367
- 2023Can We Edit Factual Knowledge by In-Context Learning?cited by 284
- 2022Aging with GRACE: Lifelong Model Editing with Discrete Key-Value Adaptorscited by 267
- 2024SafeDecoding: Defending against Jailbreak Attacks via Safety-Aware Decodingcited by 233
- 2023LLM Self Defense: By Self Examination, LLMs Know They Are Being Trickedcited by 226
- 2023Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignmentcited by 213
- 2023Transformer-Patcher: One Mistake worth One Neuroncited by 196
- 2024A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicitycited by 179
- 2024Secrets of RLHF in Large Language Models Part II: Reward Modelingcited by 170
- 2023Defending Large Language Models Against Jailbreaking Attacks Through Goal Prioritizationcited by 156
- 2024Detoxifying Large Language Models via Knowledge Editingcited by 86
- 2024Defending LLMs against Jailbreaking Attacks via Backtranslationcited by 65
- 2025Reinforcement Learning from Human Feedbackcited by 53
- 2024Knowledge Mechanisms in Large Language Models: A Survey and Perspectivecited by 50
Built on by
Citing papers per year: 2025: 1; 2026: 2.
3 papers.