SafetyLock for Large Language Models

We have released SafetyLock for the Llama3 family, Llama3.1 family, and Mistral family. When you need to fine-tune these models, you only need to add the SafetyLock specific to that model after completing the fine-tuning process to quickly restore the model's safety features.

For smaller models, we have released SafetyLock with TopK=24. For larger models, we have released SafetyLock with both TopK=24 and TopK=48. Generally, larger models perform better when using SafetyLock with a higher TopK value.