Atlas OSS: Tools for applied deep learning

Atlas OSS:

Tools for applied deep learning,

now open source

An open-source, self-hosted platform for developing applied machine learning and deep learning projects. Atlas’ SDK and CLI allow you and your team to schedule, run, track and reproduce DL jobs concurrently and share GPU & CPU resources.

How it works:

1. Import Seamlessly

Merge Atlas easily into any Python code. The lightweight SDK is flexible to use with any Python frameworks.

2. Run Experiments

Atlas’ scheduler manages resources and runs experiments from the terminal or SDK.

3. Track Results

Atlas' GUI automatically records everything about your experiments, making reproducibility a cinch.

Atlas Features

Large-scale

experimentation

Atlas’s built-in scheduler automates, queues, and executes 1000s of experiments concurrently, using any IDE or Python framework.

Experiment management & tracking

Tag experiments and easily track hyperparameters, metrics, and artifacts such as images, GIFs, and audio clips in a web-based dashboard.

Built-in integrations

Send your jobs to TensorBoard with one click using Atlas. Ramp up data processing time using NVIDIA RAPIDS with Atlas’ custom worker image functionality.

Tutorials

CIFAR Tutorial

Learn how to take advantage some of Atlas' best features.

Learn more

Atlas on AWS

Learn how to run Atlas jobs on AWS for faster results.

Learn more

Image Segmentation

Use Atlas and Tensorflow to perform image segmentation.

Learn more

Using Atlas at Square

Dessa is part of the Square ecosystem, where our team uses Atlas to build deep learning systems that empower people financially and in general, stretch what’s possible with AI.

As of June 2020, Atlas is maintained exclusively by the ML community worldwide. To contribute to Atlas as an open-source tool, check out the Github for it here.