A simple web app integrated with the Mistral model.
This project provides simple web interface for chatting with Mistral model. You can also load this model using 4/8-bit quantization, float16 and/or use flash attention.
- python3.11
- docker
- nvidia-container-toolkit
- docker compose
- Mistral instruct model from HuggingFace
In docker-compose.yml file edit volume path to model:
volumes:
- X:/Mistral-7B-Instruct-v0.2/model:/model
Build image and run compose:
docker compose build
docker compose up -d
Run in your browser:
At the start of application the model is not loaded. Choose your option and click "Load model" button.