Xiangteng NPU Chip [6] – Chip Usage Guide and Software Development Reference Materials Summary

The Xiangteng NPU chip HKN201 neural network processor is a domestically produced, high-performance, multi-core NPU processor launched by Xi’an Xiangteng Microelectronics Technology Co., Ltd., aimed at strong real-time, high-performance, and high-concurrency application scenarios.Applications of the HKN201 neural network processor chip can refer to the NPU chip inference case guidance and Linux platform application guidance documents. Application software development can refer to theXiangteng NPU processor compilation tool user manual,Xiangteng NPU chip software development manual, etc.

1. Compiler Usage Cases and Inference Experiment Cases

The Xiangteng NPU Compiler Usage Cases and Inference Experiment Cases introduces how to use the HKN201 neural network processor compiler, how to complete an inference experiment on the Xiangteng NPU baseboard, with main content including

Chapter 1 Introduction to Resources Required for Downloading the Application Compiler (1) Download the Ubuntu image of the NPU compilation environment via Baidu Cloud

• Address:https://pan.baidu.com/s/1W3grNpbyBhfO1RX3pgMAxQ
• Extraction Code:91mo

(2) Unzip the downloaded compressed package in Ubuntu, virtual machine passwordxtic_npu (3) Prepare the trained ONNX file and test image setChapter 2 Xiangteng NPU Compiler User Instructions

• First, unzip the compiler and open the compiler

• Then, import the ONNX file, and configure the parameters of the compiler

• Next, click the “Compile” button, it will display “Compiling”, and the compilation information will be printed in the compilation information area

• Finally, click “Output File”, and a bin file will be generated in the specified directory, retrieve the bin file

Chapter 3 Xiangteng NPU Development Board Inference Experiment

• Start the development board, connect WiFi, and verify the WiFi connection is successful by pinging.

• Confirm the connection status of the PCIe channel between the development board and the NPU card, and use the lspci command to confirm the PCIe device is in place.

• Copy the compiled YoloV8 bin file from the virtual machine to the local machine to complete the inference experiment. Refer to the following image to complete 4 markers and achieve inference application.

2. LINUX System Platform Usage Guidance

The HKN201 Neural Network Processor LINUX System Platform Usage Guidance details how users can use the HKN201 neural network processor on the Linux platform, including steps such as connecting the development board to a PC, achieving PCIe connection, and completing inference experiments.

3. Compilation Tool User Manual

The HKN201 Neural Network Processor Compilation Tool User Manual details the functions and usage methods of the compilation tool that accompanies the HKN201. As a bridge between neural network algorithms and HKN201, the compilation tool can convert neural network algorithm models and weight data into a format that HKN201 can recognize for algorithm configuration and weight data..

The compilation tool for the HKN201 neural network processor supports algorithm quantization method selection and configuration, multi-core acceleration configuration, input data preprocessing configuration (precision conversion method configuration, mean/normalization configuration, image input size scaling configuration), algorithm computation graph optimization, and algorithm runtime memory usage assessment, among other functions. The compilation tool is powerful and easy to use.

4. Software Development Manual

The HKN201 Neural Network Processor Software Development Manual details the API interface functions and usage methods in the HKN201 runtime software, including

• Overview of the architecture of the HKN201 runtime software.

• Describes the neural network inference management API, including inference engine management, neural network algorithm management, data management, and core resource management.

• Provides programming examples to illustrate the usage of the five working modes of the HKN201.

Appendix: HKN201 Neural Network Processor Software Development Manual

For detailed information, refer to the Hard Ten Classroom, scan the QR code to enterhttps://www.hw100k.com/coursedetail?id=271

Contact Information for Chip Consultation and Procurement