Distributask

A simple way to distribute rendering tasks across multiple machines.

Description

Distributask is a package that automatically queues, executes, and uploads the result of any task you want using Vast.ai, a decentralized network of GPUs. It works by first creating a Celery queue of the tasks, which contain the code that you want to be ran on a GPU. The tasks are then passed to the Vast.ai GPU workers using Redis as a message broker. Once a worker has completed a task, the result is uploaded to Hugging Face.

Installation

pip install distributask

Development

Setup

Clone the repository and navigate to the project directory:

git clone https://github.com/DeepAI-Research/Distributask.git
cd Distributask

Install the required packages:

pip install -r requirements.txt

Or install Distributask as a package:

pip install distributask

Configuration

Create a .env file in the root directory of your project or set environment variables to create your desired setup:

REDIS_HOST="name of your redis server"
REDIS_PORT="port of your redis server
REDIS_USER="username to login to redis server"
REDIS_PASSWORD="password to login to redis server"
VAST_API_KEY="your Vast.ai API key"
HF_TOKEN="your Hugging Face token"
HF_REPO_ID="name of your Hugging Face repository"
BROKER_POOL_LIMIT="your broker pool limit setting"

Getting Started

Running an Example Task

To run an example task and see Distributask in action, you can execute the example script provided in the project:

# Run the example task locally using either a Docker container or a Celery worker:
python -m distributask.example.local

# Run the example task on Vast.ai ("kitchen sink" example):
python -m distributask.example.distributed

This script configures the environment, registers a sample function, creates a queue of tasks, and monitors its execution on some workers.

Command Options

• --max_price is the max price (in $/hour) a node can be be rented for.
• --max_nodes is the max number of vast.ai nodes that can be rented.
• --docker_image is the name of the docker image to load to the vast.ai node.
• --module_name is the name of the Celery worker.
• --number_of_tasks is the number of example tasks that will be added to the queue and done by the workers.

Documentation

For more info checkout our in-depth documentation!

Contributing

Contributions are welcome! For any changes you would like to see, please open an issue to discuss what you would like to see changed or to change yourself.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

@misc{Distributask,
  author = {DeepAIResearch},
  title = {Distributask: a simple way to distribute rendering tasks across mulitiple machines},
  year = {2024},
  publisher = {GitHub},
  howpublished = {\url{https://github.com/DeepAI-Research/Distributask}}
}

Contributors