Step 5: Benchmark the model on devices

NetsPresso Benchmarker measures the latency of the actual device, allowing you to profile the features of models in advance.

ℹ️

Please double-check the following elements before benchmarking.

• Check the target model directory. You can verify whether the target model is a Compressed Model or a Trained Model through the directory.
• Benchmarkable models are those that have passed through the NetsPresso Convert module. If a model does not pass benchmarking, a notification will appear. When you select the device for benchmarking below, you will be automatically redirected to the page for setting up the conversion experiment.

Range of Support

The supported Hardware includes ARM, NVIDIA, RaspberryPi and Intel. Please refer to the list of supported device in the GitHub README (link). Both Python package and GUI are compatible.

Benchmark Configuration

The selectable option is Target Device(Data type).

• Drop-down: Target Device (Data Type)