Skip to content

NN lifecycle

Model can be created in different ways. This page covers all possible scenarious and describes the main storage principles.

Model storage: Supervisely

If the model stores on Supervisely Server, this means that it can be used on all computational nodes. Before task (training or inference) Agent will download model to its local storage if it doesn't exist.

Model storage: Specific Node

If the model stores on specific node in CLuster, this means that it can be used (training or inference) only on this node. If you are going to save this model or use it on other Cluster nodes, you have to upload it to Supervisely Server.

Model Zoo

If model was added to account from Model Zoo, its main store location is on Supervisely Server. This means that this model can be used on all computational nodes from Cluster.

Here is the example:

Training Checkpoint

After training user can create model from any checkpoint. This model will be added to "My models" page. This model stores on node that was used for training.

To save this model or to be able to use it on different node just click "three dots" -> "Upload to Supervisely" button.

After you click this button special task will be created in "Neural Networks" -> "Tasks". You will be redirected automatically.

After the task is completed the model is stored on Supervisely Server.

Imported model

Imported model is stored on Supervisely server by default.