TPU Benchmark app

Project description

Benchmarking and converting neural networks for different hardware platforms is a crucial task in the field of machine learning. The TPU Benchmark app is designed to facilitate this process by providing a user-friendly interface for testing and comparing the performance of neural networks across a few hardware configurations. Supported platforms include Jetson Nano, CPU and Coral Edge TPU.

An additional feature of the app is that models can be converted from TensorFlow or PyTorch to ONNX format, TF Lite, TensorRT and format compatible with Coral Edge TPU. So the tool can be used not only for benchmarking, but also for converting models to different formats.

The project can be downloaded from GitHub. The repository is private for now.

Key technologies

• Python
• Docker
• TensorFlow
• PyTorch
• ONNX
• TensorRT

Results

The project has been successfully completed.