BM1684-based AI Edge Server - Model Conversion, Large Model Integrated Machine

If you encounter any issues during the installation, you can refer to online resources for troubleshooting.

Download SDK

Next, you will need to download the necessary SDK from the Sophon official website. The SDK includes several essential tools for model compilation and deployment:

• tpu-nntc: This tool is responsible for offline compilation and optimization of neural network models trained using various deep learning frameworks, generating the BModel required for runtime execution. It supports frameworks like Caffe, Darknet, MXNet, ONNX, PyTorch, PaddlePaddle, and TensorFlow.
• libsophon: This library provides BMCV, BMRuntime, and BMLib, which are used to drive hardware like VPP and TPU for image processing, tensor operations, and model inference.
• sophon-mw: This encapsulates libraries like BM-OpenCV and BM-FFmpeg, facilitating video stream parsing and image encoding/decoding acceleration for deep learning applications.
• sophon-sail: This provides high-level interfaces in Python and C++, encapsulating lower-level library interfaces for deep learning application development.

You can download the SDK from the following links:

• tpu-nntc
• libsophon
• sophon-mw
• sophon-demo
• sophon-img
• sophon-sail

You can download these files using wget: