10.1. Model Composer Overview

Edge AI Studio Model Composer is a graphical front end for the same training and compilation flow that Tiny ML ModelZoo runs from the command line.

Note

The web UI is not part of this repository and could not be verified for the 1.5.0 guide. Treat the feature list as a general description and check the live application.

10.1.1. What it does

  • Upload and manage datasets

  • Configure training visually

  • Train models with progress visualization

  • Export trained models and compiled artifacts for deployment

10.1.2. Accessing Model Composer

Model Composer runs in the browser through Edge AI Studio: https://dev.ti.com/edgeaistudio/. You need a TI account. Use a current version of Chrome, Firefox or Edge and an Internet connection.

10.1.3. Typical features

Dataset management: upload CSV or ZIP datasets, preview data, time series plots, format validation.

Model configuration: device selection, model selection, feature extraction presets.

Training: start training, loss graphs and progress indicators.

Results: accuracy metrics, confusion matrix and other plots (the command line flow produces comparable plots, see Understanding the Run Output).

Export: trained model, compiled artifacts and a configuration file for the command line. See Exporting Models.

10.1.4. Limitations

Compared with the command line, the GUI has fewer configuration options, may not expose all presets, and has upload size and processing time limits. For custom models (Bring Your Own Model (BYOM)) or custom datasets (Bring Your Own Data) use the command line.

10.1.5. Combining the GUI and the command line

A common flow is to prototype in Model Composer, export the configuration, then refine it with the command line. The steps are in Exporting Models.

10.1.6. Next steps