Model Profiler Utility

The model profiler utility effectively wraps the Model Profiler that comes with the MLTK into a standalone executable with a webpage interface. It allows for profiling .tflite formatted models in a simulator or on a locally connected embedded target.
No installation required.


  • All model profiling is done locally. No data is uploaded to a remote server

  • Simulated numbers are based on the EFR32xG24


The model profile utility may be downloaded from here:


To use the model profiler utility:

  1. Download the executable from the link above

  2. After downloading, download-click the executable

  3. The executable will:
    1 ) Start a local HTTP server on http://localhost:8080 2 ) Open a webpage in your default browser

  4. From the webpage, drag and drop your .tflite or model file onto the box

  5. Select the hardware accelerator to use for profiling

  6. Select whether you want to profile using the simulator or locally connected embedded device

  7. Click the Profile Model button

This will profile the given model then display the results.
Refer to the Model Profiler guide for more details on the displayed profiling results.

Saving the results

After profiling completes, you can save the results in JSON or text format by clicking the respective button on the top-right:

Local HTTP server options

The profiler webpage is served from a local HTTP server.
By default, the server starts on localhost:8080.

The host and port may be changed by specifying the command line options:

mltk_model_profiler [--host <host>] [--port <port>]


  • <host> is the local network host interface IP address, e.g.

  • <port> is the HTTP server listening port, e.g. 8000