Stage 3: Model Testing

If you want to get more informations, please visit Benchmarker: Features & Scope of support

Simulator

• Simulates how the model would perform on target devices before deploying, including latency and memory footprint estimation.

📘

NetsPresso advantages:

• Simulate inference on multiple runtime environments (TensorRT, TFLite, DRPAI, ONNX Runtime, etc.)
• Quickly iterate and identify bottlenecks or compatibility issues prior to physical deployment
• Fast pre-check to optimize development workflow

Profiler

• Provides detailed performance metrics layer by layer, helping to identify bottlenecks.
• Access to a cloud-based Device Farm with 20+ diverse devices
• Receive comprehensive results (latency, FPS, memory usage) for multiple runtime environments
• Visual comparison tools to analyze performance gains from optimization

Device Farm

• A remote testbed of actual edge devices (e.g., Jetson, NXP, ARM) to run real benchmarks and verify compatibility.

📘

NetsPresso advantages:

• No need to own or manage multiple hardware boards—simply upload and select devices
• Supports benchmarking for both FP32 and INT8/quantized models
• Automated job scheduling and results retrieval for efficient QA and deployment validation