Paper
MoGU: A Framework for Enhancing Safety of Open-Sourced LLMs While Preserving Their Usability [ https://arxiv.org/abs/2405.14488 ]
Github: [ https://github.com/DYR1/MoGU/ ]
Introduction
We open-source the safer Llama-2-7b-chat-hf [ https://huggingface.co/meta-llama/Llama-2-7b-chat-hf ], trained with our proposed MoGU framework. Technical details of MoGU can be found in the paper ''MoGU: A Framework for Enhancing Safety of Open-Sourced LLMs While Preserving Their Usability''. We release the parameters and the inference code. The current inference code is still in a simple version and we will further improve it. In the future, we plan to open-source the training data and the training code.
You can use our open-sourced LLM by following the steps below.
Configuration Environment
pip install -r requirement.txt
Inference Stage
python inference.py
Some test examples are provided in inference.py.