README.md

MODEL SAFETY

Evaluate safety of local LLMs.

Build

• Install Ollama from https://ollama.com/download
• Clone the repo and cd into this directory.
• Run setup.bat

Run

• eval_model.py

python src\eval_model.py

This program reads unsafe prompts from the data directory and asks the llm to classify if they are safe or unsafe. The output is generated in the output directory. Use the -h (--help) flag for usage details.