Model Inference Compatibility Testing Method Based on Autonomous and Controllable Computing Power

As the requirements for computing platforms in deep learning model inference tasks increase, autonomous and controllable computing power platforms have become an important choice for ensuring information security and optimizing performance. This study proposes a model inference compatibility testing method based on autonomous and controllable computing power, aimed at evaluating the performance and compatibility of different autonomous and controllable computing power platforms in deep learning inference tasks. The study designs experiments such as dynamic batch response time testing, hardware resource occupancy monitoring, and multi-precision floating-point operation throughput evaluation, providing a comprehensive assessment of the inference efficiency and adaptability of different platforms. Experimental results show that the Ascend platform demonstrates high computational efficiency in large-scale inference tasks, while the Kunlun II Core platform excels in low-latency and power management. By optimizing precision settings, inference throughput can be significantly improved. This study provides a theoretical basis and practical reference for the selection and deployment of autonomous and controllable computing power platforms.