Badllama 3: removing safety finetuning from Llama 3 in minutes
Fuente:
arXiv
Gespeichert in:
| 1. Verfasser: | |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866929404973678592 |
|---|---|
| author | Volkov, Dmitrii |
| author_facet | Volkov, Dmitrii |
| contents | We show that extensive LLM safety fine-tuning is easily subverted when an attacker has access to model weights. We evaluate three state-of-the-art fine-tuning methods-QLoRA, ReFT, and Ortho-and show how algorithmic advances enable constant jailbreaking performance with cuts in FLOPs and optimisation power. We strip safety fine-tuning from Llama 3 8B in one minute and Llama 3 70B in 30 minutes on a single GPU, and sketch ways to reduce this further. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_01376 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Badllama 3: removing safety finetuning from Llama 3 in minutes Volkov, Dmitrii Machine Learning Artificial Intelligence Computation and Language Cryptography and Security We show that extensive LLM safety fine-tuning is easily subverted when an attacker has access to model weights. We evaluate three state-of-the-art fine-tuning methods-QLoRA, ReFT, and Ortho-and show how algorithmic advances enable constant jailbreaking performance with cuts in FLOPs and optimisation power. We strip safety fine-tuning from Llama 3 8B in one minute and Llama 3 70B in 30 minutes on a single GPU, and sketch ways to reduce this further. |
| title | Badllama 3: removing safety finetuning from Llama 3 in minutes |
| topic | Machine Learning Artificial Intelligence Computation and Language Cryptography and Security |
| url | https://arxiv.org/abs/2407.01376 |