ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2403.01251
31
2

Accelerating Greedy Coordinate Gradient via Probe Sampling

2 March 2024
Yiran Zhao
Wenyue Zheng
Tianle Cai
Xuan Long Do
Kenji Kawaguchi
Anirudh Goyal
Michael Shieh
ArXivPDFHTML
Abstract

Safety of Large Language Models (LLMs) has become a critical issue given their rapid progresses. Greedy Coordinate Gradient (GCG) is shown to be effective in constructing adversarial prompts to break the aligned LLMs, but optimization of GCG is time-consuming. To reduce the time cost of GCG and enable more comprehensive studies of LLM safety, in this work, we study a new algorithm called Probe sampling\texttt{Probe sampling}Probe sampling. At the core of the algorithm is a mechanism that dynamically determines how similar a smaller draft model's predictions are to the target model's predictions for prompt candidates. When the target model is similar to the draft model, we rely heavily on the draft model to filter out a large number of potential prompt candidates. Probe sampling achieves up to 5.65.65.6 times speedup using Llama2-7b-chat and leads to equal or improved attack success rate (ASR) on the AdvBench. Furthermore, probe sampling is also able to accelerate other prompt optimization techniques and adversarial methods, leading to acceleration of 1.8×1.8\times1.8× for AutoPrompt, 2.4×2.4\times2.4× for APE and 2.4×2.4\times2.4× for AutoDAN.

View on arXiv
Comments on this paper