AI Paper Atlas

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

SPQR: A Multi-Dimensional Benchmark for Safety Alignment under Benign Model Adaptation

Mohammed Talha Alam, Nada Saadi, Fahad Shamshad, Nils Lukas, Karthik Nandakumar, Fakhri Karray, Samuele Poppi

ECCV 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 4 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
fine-tune with LoRA instead of full fine-tuning (ablation)SalUn · SD1.5 benign fine-tuningrobustness R: 0.726 -> 1.000Improved
fine-tune with LoRA instead of full fine-tuning (ablation)ESD · SD1.5 benign fine-tuningrobustness R: 0.684 -> 0.950Improved
fine-tune with LoRA instead of full fine-tuning (ablation)AdvUnlearn · SD1.5 benign fine-tuningrobustness R: 0.159 -> 0.087Hurt
fine-tune with LoRA instead of full fine-tuning (ablation)SPM · SD1.5 benign fine-tuningrobustness R: 0.684 -> 0.571Hurt
fine-tune with LoRA instead of full fine-tuning (ablation)UCE · SD1.5 benign fine-tuningrobustness R: 0.942 -> 0.819Hurt
fine-tune cross-attention only instead of full UNet (ablation)RECE · SD1.5 benign fine-tuningrobustness R: 0.980 -> 0.942No clear effect
switch benign fine-tuning data from English general to multilingualESD · SD1.5robustness R: 0.684 -> 0.291Hurt
switch benign fine-tuning data from English general to ArabicMACE · SD1.5robustness R: 0.657 -> 0.979Improved
evaluate on adversarial RAB prompts instead of ViSUESD · SD1.5 standard profilerobustness R: 0.684 -> 0.020Hurt
evaluate RECE built on UCERECE vs UCE · SD1.5 generalrobustness R: 0.942 -> 0.980Improved
benign fine-tuning raises utility while safety dropsEraseDiff · SD1.5delta CLIP and delta FID: +0.111, -216.764Improved
use raw CLIPScore for prompt adherence normalization (ablation)SPQR score · ranking stabilitySpearman rho: 1.000No clear effect

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