Multi-Turn Distributed Prompt Injection Detection: Dataset and Model Weights

Synthetic shared-prefix multi-turn prompt injection conversations and trained model weights (dual-encoder GRU+LSTM, attention variant, DistilBERT baselines, ablation checkpoints) supporting the paper 'Multi-Turn Distributed Prompt Injection Detection'. Anonymized for double-blind peer review.

本数据集包含支撑论文《多轮分布式提示词注入（prompt injection）检测》（Multi-Turn Distributed Prompt Injection Detection）的合成共享前缀多轮提示词注入对话数据与训练完成的模型权重，其中涵盖双编码器GRU+LSTM、注意力变体、蒸馏BERT（DistilBERT）基线模型及消融实验检查点。本数据集已完成匿名化处理，以满足双盲同行评审的要求。