| --- |
| license: apache-2.0 |
| datasets: |
| - HPAI-BSC/Egida |
| language: |
| - en |
| base_model: |
| - Qwen/Qwen2.5-7B-Instruct |
| tags: |
| - safety |
| --- |
| |
| <div align="center" style="line-height: 1;"> |
| <a href="https://arxiv.org/abs/2502.13603" target="_blank" style="margin: 2px;"> |
| <img alt="Paper" src="https://img.shields.io/badge/arXiv-2502.13603-b31b1b.svg" style="display: inline-block; vertical-align: middle;"/> |
| </a> |
| <a href="https://huggingface.co/collections/HPAI-BSC/egida-llm-safety-67b5b15d12bc9887d0045598" target="_blank" style="margin: 2px;"> |
| <img alt="Egida Collection" src="https://img.shields.io/badge/Egida_Collection-Hugging%20Face-FFD21E?logo=huggingface" style="display: inline-block; vertical-align: middle;"/> |
| </a> |
| <a href="https://hpai.bsc.es/" target="_blank" style="margin: 2px;"> |
| <img alt="HPAI Website" src="https://img.shields.io/badge/HPAI-Website-blue" style="display: inline-block; vertical-align: middle;"/> |
| </a> |
| <a href="https://www.linkedin.com/company/hpai" target="_blank" style="margin: 2px;"> |
| <img alt="LinkedIn" src="https://custom-icon-badges.demolab.com/badge/LinkedIn-0A66C2?logo=linkedin-white&logoColor=fff" style="display: inline-block; vertical-align: middle;"/> |
| </a> |
| <a href="https://bsky.app/profile/hpai.bsky.social" target="_blank" style="margin: 2px;"> |
| <img alt="Bluesky" src="https://img.shields.io/badge/Bluesky-0285FF?logo=bluesky&logoColor=fff" style="display: inline-block; vertical-align: middle;"/> |
| </a> |
| </div> |
| |
| ## Model Description |
|
|
| - **Fine-Tuned from Model:** [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) |
| - **Paper:** [Efficient Safety Retrofitting Against Jailbreaking for LLMs](https://arxiv.org/abs/2502.13603) |
| - **Point of Contact:** [Adrián Tormos](mailto:adrian.tormos@bsc.es) |
|
|
|
|
| ## Model Summary |
|
|
| This is a fine-tuned Qwen2.5-7B-Instruct model on the [Egida-DPO-Qwen2.5-7B-Instruct](https://huggingface.co/datasets/HPAI-BSC/Egida/viewer/Egida-DPO-Qwen2.5-7B-Instruct) dataset. |
|
|
| The [Egida](https://huggingface.co/datasets/HPAI-BSC/Egida/viewer/Egida?views%5B%5D=egida_full) dataset is a collection of adversarial prompts that are thought to ellicit unsafe behaviors from language models. Specifically for this case, the Egida train split is used to run inference on Qwen2.5-7B-Instruct. Unsafe answers are selected, and paired with safe answers to create a customized DPO |
| dataset for this model. This results in a DPO dataset composed by triplets < ”question”, ”chosen answer”, ”discarded answer” > which contain questions that elicit unsafe responses by this target model, as well as the unsafe responses produced by it. |
|
|
| ## Training Details |
|
|
| - **Hardware:** NVIDIA H100 64 GB GPUs |
| - **Devices:** 4 GPUs (1 node) |
| - **Time:** 1.59h |
| - **Batch Size:** 8 |
| - **LR:** 10−7 |
|
|
| ## Performance |
|
|
| ### Safety Performance (Attack Success Ratio) |
|
|
| | | Egida (test) ↓ | DELPHI ↓ | Alert-Base ↓ | Alert-Adv ↓ | |
| |------------------------------|:--------------:|:--------:|:------------:|:-----------:| |
| | Qwen-2.5-7B-Instruct | 0.471 | 0.138 | 0.544 | 0.080 | |
| | Qwen-2.5-7B-Instruct-Egida-DPO | 0.322 | 0.118 | 0.410 | 0.045 | |
|
|
| ### General Purpose Performance |
|
|
| | | OpenLLM Leaderboard (Average) ↑ | MMLU Generative (ROUGE1) ↑ | |
| |------------------------------|:---------------------:|:---------------:| |
| | Qwen-2.5-7B-Instruct | 0.488 | 0.331 | |
| | Qwen-2.5-7B-Instruct-Egida-DPO | 0.488 | 0.296 | |
|
|
| ### Refusal Ratio |
|
|
| | | OR Bench 80K (refusal) ↓ | OR Bench Hard (refusal) ↓ | |
| |------------------------------|:---------------------:|:---------------:| |
| | Qwen-2.5-7B-Instruct | 0.021 | 0.175 | |
| | Qwen-2.5-7B-Instruct-Egida-DPO | 0.029 | 0.240 | |
|
|
| Note that this refusal ratio is computed as keyword matching with a curated list of keywords. For more information, check the paper. |
|
|
|
|
| ## Environmental Impact |
|
|
|
|
| ## Citation Information |
|
|
|
|
| ``` |
| @misc{garciagasulla2025efficientsafetyretrofittingjailbreaking, |
| title={Efficient Safety Retrofitting Against Jailbreaking for LLMs}, |
| author={Dario Garcia-Gasulla and Adrian Tormos and Anna Arias-Duart and Daniel Hinjos and Oscar Molina-Sedano and Ashwin Kumar Gururajan and Maria Eugenia Cardello}, |
| year={2025}, |
| eprint={2502.13603}, |
| archivePrefix={arXiv}, |
| primaryClass={cs.CL}, |
| url={https://arxiv.org/abs/2502.13603}, |
| } |
| ``` |
