The AI ​​era has arrived. From large models to small ones, it has penetrated every industry. With it comes a wide variety of AI accelerators, including PCIe, M.2, and SoC models. Currently, mainstream inference is based on Python, but Python's efficiency is not ideal, especially in cost-sensitive areas that require real-time processing, such as video processing.
To enable AI model engineers to focus on the model, we have launched AI Container. AI Container is completely based on C++ and compatible with AI accelerator APIs from multiple manufacturers.
AI model engineers train their models and submit them directly to AI Container, which then handles subsequent inference and rule generation.
AI Container supports mainstream models such as Yolo, Arcface, and LPRnet. AI model engineers can replace models online and verify their performance at any time.
AI Container supports embedded SoC deployment, significantly reducing the cost of AI deployment.
Both the USC security platform and the HPro AI NVR will support AI Container. The first phase supports PCIE cards including NVIDIA GPU/Atlas 300V/BM1684x PCIE, supported M.2 acceleration modules including Hailo8 M.2/AX650n M.2/AX8850 M.2, and supported SOCs including BM1688 SOC/RK3588 SOC/AX650n SOC.Unified AI Container