When reviewing a solution, the interface provides detailed specifications for the selected model:
- Target metric — Displays the current metric value (for example, Validation Accuracy). You can use the Metric type dropdown to switch between Training Metrics and Validation Metrics. If a separate holdout dataset was not provided before training, the validation metrics are automatically calculated based on a 20% split of your training data.
- Total footprint — Shows the estimated SRAM and NVM memory usage. This includes a detailed breakdown for the model, the inference engine, and signal processing, calculated specifically for your selected target hardware.