Deep learning platform Atlas is now open source
The deep learning company Dessa has open sourced Atlas, a deep learning platform. Though currently still in beta mode, it is designed to make running, evaluating and deploying deep learning projects easier. It works on macOS, Linux and Windows, and offers TensorBoard integration. Let’s take a closer look.
Let’s see what features this beta version of their deep learning platform has to offer.
Features of Atlas
Atlas has a Python SDK, CLI, GUI and scheduler on board. These aspects should help reduce the effort of managing infrastructure and increase the speed of developing deep learning models.
The deep learning platform is self-hosted and runs on either a single node, a multi-node cluster or on multi-cloud clusters. It supports job scheduling to allow collaboration, and every job is reproducible because it is recorded and tracked.
On Twitter, Dessa posted a demonstration of Atlas:
We ❤️ open source, and now we hope you’ll ❤️open source Atlas:
As of today, our DL dev platform Atlas is now open source! Excited to collaborate with bright people around the world to make these tools even more helpful. Learn more here: https://t.co/OGIM5G1ZxC pic.twitter.com/LfMzJ3T0y8
— Dessa (@dessa) March 17, 2020
Installation guides for Atlas are available for macOS, Linux and Windows 10.
On Windows, for example, the installation should take 10 minutes. The requirements are a Docker version higher than 18.09, Python 3.6 or higher, more than 5GB of free storage and the atlas_installer.py file, which you can get here.
See the Square blog post for further details.