AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

EVA: Editing for Versatile Alignment Against Jailbreaks

Yi Wang, Hongye Qiu, Yue Xu, Sibei Yang, Zhan Qin, Minlie Huang, Wenjie Wang

TPAMI 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
apply EVA (text) editingVicuna-7B-v1.5 · HarmBench AutoDANASR: 0.69 -> 0.04Improved
apply EVA to a VLMLLaVA-1.5-7B · ADV-16 on HarmBenchASR: 0.71 -> 0.00Improved
apply EVA to a VLMLLaVA-1.5-7B · MultiTrust CrossmodalASR: 0.88 -> 0.00Improved
apply EVA to a VLMQwen2.5-VL-7B · MM-SafetyBench SD+TYPOASR: 0.51 -> 0.00Improved
add visual-token editing to text-only EVA (ablation)LLaVA-1.5-7B · MultiTrust MultimodalASR: 0.42 -> 0.13Improved
add visual-token editing to text-only EVA (ablation)LLaVA-1.5-7B · MMStarscore: 27.3 -> 32.3Improved
adaptive attacks regenerated on edited modelEVA (text) · Vicuna-7B-v1.5 avg ASRASR: 0.068 -> 0.112Hurt
adaptive attacks regenerated on edited modelEVA (text) · Qwen2.5-7B avg ASRASR: 0.012 -> 0.113Hurt
adaptive ADV-16 attack on edited VLMEVA · LLaVA-1.5-7B ADV-16 avg ASRASR: 0.000 -> 0.125Hurt
select image token randomly instead of by attention (ablation)EVA · Qwen2.5-VL-7B MM-Vet-v2score: 55.5 -> 10.4Hurt
select first (fixed) image token instead of by attention (ablation)EVA · Qwen2.5-VL-7B MM-Vet-v2score: 55.5 -> 12.1Hurt
use attention-based token selection rather than L2 distance (ablation)EVA · Qwen2.5-VL-7B MM-SafetyBench SD+TYPOASR: 0.08 -> 0.00Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy