Lune

ACL Findings2025

Dynamic Evil Score-Guided Decoding: An Efficient Decoding Framework For Red-Team Model

Cong Gao, Bo Zhang, Linkang Yang, Minghao Hu, Zhunchen Luo, Xiaoying Bai, Guotong Geng, Jun Zhang, Yunhua Xue

Abstract

Large language models (LLMs) have achieved significant advances, but can potentially generate harmful content such as social biases, extremism, and misinformation. Red teaming is a promising approach to enhance model safety by generating adversarial prompts to test and improve model robustness. However, existing red-teaming methods often require expensive fine-tuning, especially for large LLMs. We propose the Dynamic Evil Score-Guided Decoding framework (DESGD), an efficient red-teaming method that does not increase computational cost with the target model size. DESGD framework introduces the concept of an 'evil score,' which can dynamically evaluate the potential of tokens to contribute to harmful outputs in decoding phrases. We fine-tune a small model using an adversarial dataset and calculate the evil score based on the difference in logits vector before and after fine-tuning. Then, we adjust the logits of the large target model according to the evil score. The results of the experiment show that our method achieves an ASR of 92. 83% on the Llama-3.2-3B-Instruct model, compared to 83.48% with adversarial fine-tuning while using lower computational resources. Similarly, on the Qwen2.5-3B-Instruct model, DESGD reaches an ASR of 88.62%, outperforming adversarial fine-tuning (77.56%).

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

Builds onCited byLeft to right by year; the more cited, the higher.

Builds on8

  1. LoRA: Low-Rank Adaptation of Large Language ModelsEdward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu et al.ICLR 2022 · 18,833 citations
  2. RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language ModelsSamuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi et al.EMNLP Findings 2020 · 692 citations
  3. "Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language ModelsXinyue Shen, Zeyuan Chen, Michael Backes, Yun Shen et al.CCS 2024 · 132 citations
  4. Adversarial training for high-stakes reliabilityDaniel M. Ziegler, Seraphina Nix, Lawrence Chan, Tim Bauman et al.NeurIPS 2022 · 79 citations
  5. Weak-to-Strong Search: Align Large Language Models via Searching over Small Language ModelsZhanhui Zhou, Zhixuan Liu, Jie Liu, Zhichen Dong et al.NeurIPS 2024 · 50 citations

Related papers

  1. Be a Multitude to Itself: A Prompt Evolution Framework for Red TeamingRui Li, Peiyi Wang, Jingyuan Ma, Di Zhang et al.EMNLP Findings 2024
  2. Attack Prompt Generation for Red Teaming and Defending Large Language ModelsBoyi Deng, Wenjie Wang, Fuli Feng, Yang Deng et al.EMNLP Findings 2023 · 35 citations
  3. XGUARD: A Graded Benchmark for Evaluating Safety Failures of Large Language Models on Extremist ContentVadivel Abishethvarman, Bhavik Chandna, Pratik Jalan, Usman NaseemACL Findings 2026
  4. AdvPrompter: Fast Adaptive Adversarial Prompting for LLMsAnselm Paulus, Arman Zharmagambetov, Chuan Guo, Brandon Amos et al.ICML 2025
  5. Safety Alignment of Large Language Models via Contrasting Safe and Harmful DistributionsXiaoyun Zhang, Zhengyue Zhao, Wenxuan Shi, Kaidi Xu et al.AAAI 2026 · 4 citations
Dusk over the sea between two cliffs drawn in fine vertical lines