After clicking Next, the platform redirects you to the Model Training Tab, where you can set the model settings and monitor training progress.
-
Click New Session to proceed. This example uses the Neuton framework, where only a single session is available. When using the LiteRT framework for Axon NPU, multiple sessions can be created, allowing you to experiment with different configurations and compare the results.
-
Enter the session name. The training framework is Neuton, the created model is for Cortex-M CPU.
-
Set up the following model settings:
- Weights & Coefficients — Select Quantization-Aware 16-bit Integer. This option reduces memory footprint and speeds up inference while keeping accuracy high.
- Output Format — Select floating-point 32-bit probabilities with values from 0 to 1. This provides easy-to-read confidence scores for each gesture class.
- Early stopping — This setting is optional. Edge AI Lab will stop training when it achieving the best possible accuracy. For demonstration purposes, set a target accuracy value to stop training early. During training, Edge AI Lab performs validation. When the accuracy score reaches 0.99, training stops.