A small* Python utility for generating a password based on a user-provided input using a Large Language Model (LLM). This project is inspired by Randall Munroe's XKCD comic that suggests using four random words (no space), with a mix of nouns, verbs, adjectives, and adverbs.
The idea was to use a LLM as a sort of password manager. The LLMs output can be set to deterministic for a given seed. The seed would then be based on the site's url, so each site got their own password in a deterministic manner. This way if you have access to the same LLM and can set the seed you can then generate the same password. However, to make things a bit more secure a salt was added to the seed. Making it less portable then what was initially planned.
- Generates a password based on a user-provided input using an LLM.
- Uses a salt for additional security.
- Supports benchmarking mode for generating passwords for a list of sites.
- Clone the repository:
git clone https://github.com/your-username/llama-password-gen.git
cd llama-password-gen- Install the required dependencies:
pip install -r requirements.txt
- Make sure you have Ollama installed and running locally. You can download it from here.
- Pull the Llama3:8b model using
ollama pull llama3:8b
To generate a password based on a user-provided input, run the following command:
python main.py --input "your-input-here"
To run the benchmark mode and generate passwords for a list of sites, create a sites.txt file with one site per line and run the following command:
python main.py --benchmark
* not including the LLM