Skip to main content

Visualize a model

Open a model file and inspect its layers.

This guide uses an ONNX model to show how to explore a network and inspect the tensors passed between its layers. Have a .onnx file ready on your computer. The screenshots use MobileNet V2; you can follow the same steps with your own model.

Import the model#

Select Visualizer in the Hub sidebar to open Add Models.

Choose your file with + ONNX (1). Once it is Prepared (2), select Visualize (3).

For a model stored as a run artifact, choose Visualize from its file actions.

Find a layer#

Use Blocks to see the main parts of the network, then switch to individual operations to inspect a layer. Pan and zoom through the graph, or search by name.

Operations (1) shows individual layers. Use Search nodes (2) to find a layer and select it to open Node details (3).

Inspect the tensors#

Look at the selected layer’s input and output shapes to see how it changes the data. For a convolution, compare the channel count and spatial dimensions. Select a connection in the graph to inspect the tensor passed to the next operation.

Use the precision information to check the types recorded in the model. When inspecting a quantized model, look for where those types change between operations.

Save the workspace#

Workspaces are temporary. An exported workspace preserves your inspection so you can continue later or share it with a teammate.

Use Download .embedl bundle in the toolbar to save a .embedl file. Reopen it with Open .embedl in Add Models, or upload it as a run artifact to share it with your team.

Next steps#

Add an exported or compiled version to compare model stages and follow the corresponding layers between them. Use Experiment Tracking to compare their measured accuracy and hardware performance.