AI Paper Atlas

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

Toward Safe Quantization-Aware Fine-tuning: Understanding and Mitigating Safety Alignment Degradation

Yuning Yang, Guowei Peng, Xiurui Xie, Minrui Jiang, Shuang Liang, Guisong Liu

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning

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
quantize LoRA to 4-bit (QLoRA)LoRA fine-tuning · SST5, Llama-3-8B-InstructHarmful Score: +8.2 Harmful Score, about 0.9 lower Finetune AccuracyHurt
quantize LoRA to 4-bit (QLoRA)LoRA fine-tuning · Llama-3-8B-Instruct linear probeprobe accuracy drop: 1.73% -> 3.52%Hurt
quantize LoRA to 4-bit (QLoRA)LoRA fine-tuning · VISAGE safety metricVISAGE drop: 6.34 -> 13.6Hurt
add safety-aware initialization and refinement (ExSQF) to QLoRAQLoRA · SST5, avg over poison ratiosHarmful Score: 52.3 -> 14.4Improved
add safety matrix projection to LoftQ initializationLoftQ · SST5, avg over poison ratiosHarmful Score: 29.7 -> 14.4Improved
initialize adapters from principal components of pre-trained weights (QPiSSA) instead of QLoRAQLoRA · SST5, avg over poison ratiosHarmful Score: 52.3 -> 70.4Hurt
initialize adapters from quantization error (LoftQ) instead of QLoRAQLoRA · SST5, avg over poison ratiosHarmful Score: 52.3 -> 29.7Improved
apply safety projection on output matrix O only versus Q, K, or V (ablation)ExSQF refinement · SST5Harmful Score: 15.8 -> 12.6Improved
apply safety projection to all of Q/K/V/O instead of O only (ablation)ExSQF refinement · SST5Harmful Score: 12.6 -> 15.8Hurt
transfer SafeLoRA and AsFT to quantized models (ablation)full-precision safety methods · SST5 poison ratiosHarmful ScoreMixed
use ExSQF on GSM8K fine-tuningExSQF · GSM8KFinetune Accuracy: 69.4 -> 75.0Improved
use ExSQF at high learning ratesExSQF · SST5 learning rate sweepHarmful ScoreImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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