A easy-to-use system to allow every data scientist to have access to power hardware. With out-of-the-box feature, all you need to do is install it on your cluster/s and start melting your GPUs 🎉
Dora is an open source and free distributed system for managing AI workloads easy. It's composed by a control plane (with an API, a scheduler and other services), by compute nodes, and by clients. Available clients are: CLI, Web App, Electron App. Linux, MacOS and Windows are supported.
BETA
Dora is in beta stage, release 0.8.1 🎉 💯
With Dora you can setup and run complex AI experiments (but not only) with one command:
# my.experiment.yaml
---
apiVersion: v1
kind: Workload
metadata:
name: my.experiment
spec:
replica:
count: 4
driver: Docker
selectors:
gpu:
product_name:
- Tesla V100-SXM2-16GB
count: 8
image:
image: tensorflow/tensorflow:lates-gpu
config:
cmd: python3 experiment.py
dora apply -f my.experiment.yaml
Then you will have your workload running on 4 nodes each with 8 GPUs, in less then 5 seconds.
If you don't want to mess up with shell commands, Dora has a beatiful, customizable web based UI
With a pull request, an issue... or fork it and start develop your own version!
Star it! GitHub (opens new window)