Foundations Atlas: 

Run, manage & evaluate 1000s of experiments using any IDE or python framework. 

Seamless integration with Tensorboard & NVIDIA's RAPIDS framework.

Lower your GPU compute costs by over 50% with Atlas' job scheduler.

a machine learning tool for high-performing data scientists

Over 1000 ML practitioners love and trust Atlas

Foundations makes the frustrating parts of machine learning easy and even enjoyable, allowing you to have insights you wouldn’t have otherwise.

Creator of AlexNet
Alex Krizhevsky,

Foundations allowed our team to quickly build NLP functionality into our core product, ensuring  robust and accurate insights for our customers.

Laura McGee,
CEO of Diversio

Your Code Here

Track metrics, params 

and artifacts

Deploy jobs from the terminal or a script

It all starts with an import...

Merge Atlas seamlessly into any Python code

The Atlas python SDK is flexible to use with any IDE or frameworks

Atlas’s built-in scheduler automatically handles the queuing and execution of jobs

Track and analyze your results in a dashboard

The SDK, Dashboard, and scheduler all run on your own machine so your data never leaves your machine.

Large-scale 
experimentation 

Rapidly queue and run 1000s of experiments concurrently on whatever hardware you’re working with.

 

The Atlas scheduler can run on cloud, an on-premise cluster, or on the machine under your desk. 

Built-in integrations

Send your jobs to TensorBoard with one click or easily improve your data processing speed  with NVIDIA RAPIDS framework using Atlas's custom worker image functionality.

Your workflow, your rules

Atlas is built to be non-intrusive, allowing you to run almost any ordinary Python code without modification.

 

View artifacts as your models are training to check their progress.

Built with support from the best in the industry

Download  Version 2.0

 

Stay tuned for more from Dessa.

ATLAS, DESSA, the Dessa Logo, and others are trademarks of Square, Inc. All third party names and trademarks are properties of their respective owners and are used for identification purposes only.