Research map / Large language models / Instruction and fine-tuning
Sysformer: Safeguarding Frozen Large Language Models with Adaptive System Prompts
Kartik Sharma, Yiqiao Jin, Vineeth Rakesh, Yingtong Dou, Menghai Pan, Mahashweta Das, Srijan Kumar
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add Sysformer adaptive system prompt to frozen LLM vs default system prompt | Llama-3.1-8B · JailbreakBench (test split) | refusal gap: 0.7000 -> 0.9334 | Improved |
| Add Sysformer vs default system prompt | Phi-3.5-mini · JailbreakBench (test split) | refusal gap: 0.0667 -> 0.7000 | Improved |
| Use Sysformer instead of static SystemEmbedder | Llama-2-7b-chat · JailbreakBench (test split) | refusal gap: 0.4333 -> 0.8333 | Improved |
| Use Sysformer instead of LoRA fine-tuning of all layers | Llama-2-7b-chat · JailbreakBench (test split) | refusal gap: 0.8667 vs 0.8333 | No clear effect |
| Train Sysformer on JailbreakBench, test on StrongReject | Llama-3.1-8b · StrongReject | refusal gap: 0.9330 | Improved |
| Train Sysformer without jailbreak examples (ablation) | Sysformer · 16 jailbreak attacks | refusal rate | No clear effect |
| Augment training with 6 of 16 jailbreak attacks (Sysformer+JB) (ablation) | Sysformer · JailbreakBench jailbreaks | refusal rate | Improved |
| Add Sysformer on Llama-2-7B-chat | Llama-2-7B-chat · Alpaca | BERTScore: 0.8487 -> 0.8414 | Hurt |
| Add Sysformer on Llama-3.1-8B | Llama-3.1-8B · Alpaca | BERTScore: 0.8327 -> 0.8467 | Improved |
| Use intermediate compliance weight 0.2-0.5 instead of 1.0 (ablation) | Sysformer · JailbreakBench | refusal rates | Improved |
| Enable reconstruction loss (ablation) | Sysformer · JailbreakBench | refusal rates | Mixed |
| Use self-compliance responses (ablation) | Sysformer · JailbreakBench | safe refusal rate | Mixed |
Closest papers by meaning
- On Prompt-Driven Safeguarding for Large Language Models (ICML 2024 · similarity 0.90)
- UpSafe℃: Upcycling for Controllable Safety in Large Language Models (NeurIPS 2026 · similarity 0.90)
- RSafe: Incentivizing proactive reasoning to build robust and adaptive LLM safeguards (NeurIPS 2025 · similarity 0.90)
- Jailbreak Antidote: Runtime Safety-Utility Balance via Sparse Representation Adjustment in Large Language Models (ICLR 2025 · similarity 0.90)
- On Evaluating the Durability of Safeguards for Open-Weight LLMs (ICLR 2025 · similarity 0.89)
- SG-Bench: Evaluating LLM Safety Generalization Across Diverse Tasks and Prompt Types (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
