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
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 GradShield FIHS filtering | Llama-3.1-8B-Instruct finetuning · LATharm | ASR: 0.98 -> 0.01 | Improved |
| add GradShield FIHS filtering | Llama-3.1-8B-Instruct finetuning · RTA | ASR: 0.16 -> 0.06 | Improved |
| add GradShield FIHS filtering | Llama-3.1-8B-Instruct finetuning · Identity-shift | ASR: 0.75 -> 0.01 | Improved |
| replace OpenAI Moderation filter with GradShield on implicit harm | Identity-shift finetuning · Identity-shift | ASR: 0.29 -> 0.01 | Improved |
| add GradShield filtering | Llama-3.1-8B-Instruct finetuning · GSM8k + LATharm | ASR: 0.97 -> 0.00 | Improved |
| add GradShield filtering | Llama-2-7b-chat finetuning · LATharm | ASR: 0.99 -> 0.00 | Improved |
| add GradShield filtering | Qwen2.5-7B-Instruct finetuning · LATharm | ASR: 0.96 -> 0.00 | Improved |
| add second round of threshold search (ablation) | GradShield · 90% harmful ratio | ASR: 0.95 -> 0.25 | Improved |
| add second round of threshold search (ablation) | GradShield · 70% harmful ratio | ASR: 0.45 -> 0.00 | Improved |
| add second round of threshold search (ablation) | GradShield · 50% harmful ratio | ASR: 0.26 -> 0.01 | Improved |
| add GradShield filtering | Llama-3.1-8B-Instruct finetuning · LATharm utility | utility: 0.53 -> 0.53 | No clear effect |
| add GradShield on harmful data prefixed with 'I can help with this.' | Llama-3.1-8B-Instruct · prefix-evasion attack | ASR: 1.00 -> 0.07 | Improved |
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.
- Safe Lora — Safe LoRA: The Silver Lining of Reducing Safety Risks when Finetuning Large Language Models (NeurIPS 2024) · better on 3 of 3 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
- Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning Attack (ICML 2025 · similarity 0.90)
- Booster: Tackling Harmful Fine-tuning for Large Language Models via Attenuating Harmful Perturbation (ICLR 2025 · similarity 0.90)
- Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! (ICLR 2024 · similarity 0.89)
- Navigating the Safety Landscape: Measuring Risks in Finetuning Large Language Models (NeurIPS 2024 · similarity 0.89)
- Safe LoRA: The Silver Lining of Reducing Safety Risks when Finetuning Large Language Models (NeurIPS 2024 · similarity 0.89)
- Safety Alignment Should be Made More Than Just a Few Tokens Deep (ICLR 2025 · 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.
