fuegovic / watch-hf

A webhook server that automates the update and restart of Hugging Face Spaces.

watch-hf 🤗👀

A webhook server that automates the update and restart of Hugging Face Spaces.

Introduction

This project was created to simplify the deployment and maintenance of Hugging Face Spaces. This webhook server does two things:

• When it receives a notification from Watchtower, a service that monitors the updates of Docker images, it updates a Hugging Face Space by performing a "factory reboot", which resets the Space and pulls the latest docker image.
• When it receives a notification from Uptime Kuma, a service that monitors the availability of websites, stating that the Hugging Face Space is down, it tries to reboot it every 2 minutes, until Uptime Kuma sends a notification stating it's back up.

This project is built with Python, Flask, and Docker.

Installation and Usage

To install and run this project, you need to have Python 3.8 or higher, Docker, and Docker Compose installed on your machine. You also need to have a Hugging Face account and a Hugging Face Space that you want to automate.

Follow these steps to set up the project:

• Clone this repository and navigate to the project directory.

• Create a virtual environment and activate it.

• Install the required packages with pip install -r requirements.txt.

• Create a .env file and add the following environment variables:

  • SERVER_PORT: the port you want to use for the server. Default port is 5000
  • DEBUG: Flask debug, true or false. Default is false
  • IMAGE_NAME: the docker image that Watchtower is watching
  • USERNAME: your Hugging Face username. You can find your username on your profile page here.
  • TOKEN: your Hugging Face token. Use a token with write access. You can generate a token from your settings page here.
  • OWNER: the owner of the Hugging Face Space.
  • NAME: the name of the Hugging Face Space

• Run python watch.py to start the webhook server and the services.

• Configure Watchtower and Uptime Kuma to send notifications to the webhook server, using the URL http://localhost:5000/webhook.

License

This project is licensed under the MIT License. See the LICENSE file for more details.

Acknowledgements

This project uses the following resources and services:

• Hugging Face, a platform for natural language processing and deep learning.
• Watchtower, a service that monitors the updates of Docker images.
• Uptime Kuma, a service that monitors the availability of websites.
• Flask, a micro web framework for Python.
• Docker, a platform for building and running containerized applications.
• Docker Compose, a tool for defining and running multi-container applications with Docker.