Defensive M2S: Training Guardrail Models on Compressed Multi-turn Conversations
This addresses efficiency bottlenecks for deploying safety guardrails in LLM applications, though it appears incremental as it applies compression techniques to an existing problem.
The paper tackles the high computational cost of training guardrail models on full multi-turn conversation histories by proposing Defensive M2S, which fine-tunes models on compressed single-turn versions instead. This approach reduces training tokens by 93× and achieves 93.8% attack detection recall with a 94.6% reduction in inference tokens.
Guardrail models are essential for ensuring the safety of Large Language Model (LLM) deployments, but processing full multi-turn conversation histories incurs significant computational cost. We propose Defensive M2S, a training paradigm that fine-tunes guardrail models on Multi-turn to Single-turn (M2S) compressed conversations rather than complete dialogue histories. We provide a formal complexity analysis showing that M2S reduces training cost from $O(n^2)$ to $O(n)$ for $n$-turn conversations. Empirically, on our training dataset (779 samples, avg. 10.6 turns), M2S requires only 169K tokens compared to 15.7M tokens for the multi-turn baseline -- a 93$\times$ reduction. We evaluate Defensive M2S across three guardrail model families (LlamaGuard, Nemotron, Qwen3Guard) and three compression templates (hyphenize, numberize, pythonize) on SafeDialBench, a comprehensive multi-turn jailbreak benchmark. Our best configuration, Qwen3Guard with hyphenize compression, achieves 93.8% attack detection recall while reducing inference tokens by 94.6% (from 3,231 to 173 tokens per conversation). This represents a 38.9 percentage point improvement over the baseline while dramatically reducing both training and inference costs. Our findings demonstrate that M2S compression can serve as an effective efficiency technique for guardrail deployment, enabling scalable safety screening of long multi-turn conversations.