Overview

A lot of tasks in Supervisely (like training neural networks or running DTL) require long and complex computations. Moreover, usually it's a good idea to distribute the work between several machines.

Supervisely allows to create computational Cluster. Now users can connect their own computers or servers and use them to distribute tasks. It is easy as make a few clicks and execute one shell command.

So how does it work? Using cluster page you can generate unique command to run on your computer. It will start a new Supervisely Agent — a simple open-sourced tasks manager available as a Docker image. Agent will connect to Supervisely API and now you can run tasks on connected computer — host.

Supervisely Computational Cluster can be scaled horizontally without any limits (All limits are defined by licensing agreement). Now users can add tens of nodes and perform hundreds of tasks in parallel.

70% #

 

Private Data Center off the shelf

This technology allows to organize and administrate private data center inside organization without any special knowledge. Now data scientists within organization can easy share computational resources and perform multiple experiments simultaniously (NN training / inference, apply massive data augmentations and so on).

What kind of tasks you can run on Agents? Here are some examples: