GenSynth Documentation

Cycle Details

Accessed from the View Details icon in the table of cycles on the Overview tab, you can explore performance statistics and details produced by GenSynth for each of the layers in the network to explain the performance of the network.

Tip

You can view the cycle flyout panel by clicking the image15.png icon next to the model name.

Network Visualization
image80.png

You may scroll (holding the left mouse button and dragging) and zoom (using the scroll wheel on your mouse) to explore the different parts of the network. Clicking on the up and down arrows allows you to go directly to the top and the bottom of the network, respectively. GenSynth only shows the most meaningful network nodes it has identified with respect to network performance.

Nodes are coloured according to encoding efficiency relative to the current cycle, with darker blue indicating higher encoding efficiency.

If you click on a node, more information is available to you.

image21.png
Layer Visualization

Note

Instead of showing every connection and operation in the network, GenSynth shows you the layers at an appropriate abstraction to help you better understand how the network behaves with respect to network performance.

Each vertical bar represents a layer of the network containing trainable parameters. The height of the bar indicates the relative number of channels.

Layers in white indicate layers that GenSynth must preserve in the generated models due to the presence of unsupported operations or changing them would impact the shapes of inputs or outputs.

When the network is very large, the top control lets you zoom in and out on the network.

image36.png

Clicking on a vertical bar pops up a box containing:

  • Variable shape.

  • The number of trainable parameters.

  • The number of floating operations required to perform one inference (FLOPs).

  • The number of channels.

  • The encoding efficiency of the layer.

Note

Encoding Efficiency is a measure GenSynth generates to show how well-used a layer within a network is in achieving the current network performance level. The higher the encoding efficiency of a layer within a network, the more efficient a layer is within the network.