PapersPolicy12:06Chinese contentinLoRA-MINT:用中间表征审计 LoRA 微调模型的训练数据能判断某个样本有没有被拿去 LoRA 微调,在 Qwen、LLaMA 等四个模型上准确率超 95%。#LoRA#MINT#微调#AI安全aarXiv: DeepSeek@Gonzalo Mancera et al. (6)SourceRead laterReadWorth followingUsefulFollow LoRA