跨大模型迁移对抗性攻击修复，无需新数据即可零样本防御。
TeleLoRA: Teleporting Model-Specific Alignment Across LLMs
- 通过联合多模型激活信息训练统一的LoRA权重生成器。
- 在未见模型上实现零样本防御，攻击成功率显著降低。
- 适用于需要高效防御恶意触发的LLM部署场景。
大型语言模型（LLMs）中的后门攻击检测与修复通常依赖于特定模型的对齐数据，因为不同模型的触发机制各不相同。本文提出TeleLoRA（远程低秩适配），一种新框架，通过整合多个LLM的本地激活信息，学习一个统一的LoRA适配器权重生成器，从而在无对齐数据的情况下实现对未见过的LLM的零样本后门防御。该生成器设计为置换对称，可泛化至架构和规模各异的模型。我们优化了模型结构以提升内存效率，使得在大规模LLM上仅用极少计算资源即可训练。在多个后门攻击基准测试中，TeleLoRA有效降低了攻击成功率，同时保持了模型正常性能。
原文摘要 · Abstract (English)
Mitigating Trojans in Large Language Models (LLMs) is one of many tasks where alignment data is LLM specific, as different LLMs have different Trojan triggers and trigger behaviors to be removed. In this paper, we introduce TeleLoRA (Teleporting Low-Rank Adaptation), a novel framework that synergizes model-specific alignment data across multiple LLMs to enable zero-shot Trojan mitigation on unseen LLMs without alignment data. TeleLoRA learns a unified generator of LoRA adapter weights by leveraging local activation information across multiple LLMs. This generator is designed to be permutation symmetric to generalize across models with different architectures and sizes. We optimize the model design for memory efficiency, making it feasible to learn with large-scale LLMs with minimal computational resources. Experiments on LLM Trojan mitigation benchmarks demonstrate that TeleLoRA effectively reduces attack success rates while preserving the benign performance of the models.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。