AI Paper Atlas

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

GradShield: Alignment Preserving Finetuning

Zhanhao Hu, Xiao Huang, Patrick Mendoza, Emad A. Alghamdi, Raluca Ada Popa, David Wagner

ICLR 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
add GradShield FIHS filteringLlama-3.1-8B-Instruct finetuning · LATharmASR: 0.98 -> 0.01Improved
add GradShield FIHS filteringLlama-3.1-8B-Instruct finetuning · RTAASR: 0.16 -> 0.06Improved
add GradShield FIHS filteringLlama-3.1-8B-Instruct finetuning · Identity-shiftASR: 0.75 -> 0.01Improved
replace OpenAI Moderation filter with GradShield on implicit harmIdentity-shift finetuning · Identity-shiftASR: 0.29 -> 0.01Improved
add GradShield filteringLlama-3.1-8B-Instruct finetuning · GSM8k + LATharmASR: 0.97 -> 0.00Improved
add GradShield filteringLlama-2-7b-chat finetuning · LATharmASR: 0.99 -> 0.00Improved
add GradShield filteringQwen2.5-7B-Instruct finetuning · LATharmASR: 0.96 -> 0.00Improved
add second round of threshold search (ablation)GradShield · 90% harmful ratioASR: 0.95 -> 0.25Improved
add second round of threshold search (ablation)GradShield · 70% harmful ratioASR: 0.45 -> 0.00Improved
add second round of threshold search (ablation)GradShield · 50% harmful ratioASR: 0.26 -> 0.01Improved
add GradShield filteringLlama-3.1-8B-Instruct finetuning · LATharm utilityutility: 0.53 -> 0.53No clear effect
add GradShield on harmful data prefixed with 'I can help with this.'Llama-3.1-8B-Instruct · prefix-evasion attackASR: 1.00 -> 0.07Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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