Skip to content

Test Compiled AI Models

The models used in the example can be downloaded from:

  1. Copy the generated output directory to the rootfs on either the SD Card or TFTP directory.

  2. Power ON the board by pressing the POWER button for at least 1 second to boot up the board.

  3. Using the serial console, log in and navigate to the/usr/bin/ directory.

    cd /usr/bin

  4. Run the Model Executor application with the generated output directory containing the MERA data for all compiled models.

model-executor -e model_deployment_dir/config.yaml

This will print a log to the console containing the inference time per run and an average inference time for each model.

Model being executed: ad_large_int8
[2026-03-30 20:22:32.633] [console] [info] MERA 2.0 Runtime
Inference Run #1, inference time: 16.962957ms
...
Inference Run #10, inference time: 16.892876ms
------------------------------
           Results
Model: ad_large_int8
Average inference time: 16.94ms over 10 runs
------------------------------
Successfully saved output data for tensor #1 to model_deployment_dir/ad_large_int8/output_0.bin
Model being executed: kws_micronet_l
[2026-03-30 20:22:32.807] [console] [info] MERA 2.0 Runtime
Inference Run #1, inference time: 16.779249ms
...
Inference Run #10, inference time: 16.698418ms
------------------------------
           Results
Model: kws_micronet_l
Average inference time: 16.71ms over 10 runs
------------------------------
Successfully saved output data for tensor #1 to model_deployment_dir/kws_micronet_l/output_0.bin
Model being executed: tiny_wav2letter_int8
[2026-03-30 20:22:32.979] [console] [info] MERA 2.0 Runtime
Inference Run #1, inference time: 25.037624ms
...
Inference Run #10, inference time: 24.954540ms
------------------------------
           Results
Model: tiny_wav2letter_int8
Average inference time: 24.97ms over 10 runs
------------------------------
Successfully saved output data for tensor #1 to model_deployment_dir/tiny_wav2letter_int8/output_0.bin