The EN675 NPU provides 1.2 TOPS of compute and runs a predefined set of model architectures. The NPU SDK includes a trainer, a compiler, and a simulator, so customers can retrain these models on their own data. As long as the customer provides a labeled dataset, a model can be trained, compiled, and deployed on the EN675.

NPU SDK contents
Environment: Ubuntu 20 / 22 , Python 3.8

Trainer
- mobilenetv2
- mobilenetv2+SSD
- Deformed yolov5s
- Deformed yolov7-tiny

NPU Compiler
- Input model format: ONNX
- Compute precision: INT8
- Quantization scheme: per-channel symmetric (weights) / per-tensor asymmetric (activations)
- PTQ only (QAT not supported)

NPU Simulator
- Provided as an executable; runs on a local PC

EYENIX MODEL ZOO
EYENIX MODEL ZOO URL : https://github.com/EyenixG/eyenix_model_zoo

Supported Model
- mobilenetv2
- mobilenetv2+SSD
- Deformed yolov5s
- Deformed yolov7-tiny