Qualcomm and Tencent Hunyuan cooperate to jointly promote the terminal-side deployment of Tencent Hunyuan large model based on Snapdragon 8 Extreme Edition

During the Snapdragon Summit, Qualcomm Technologies announced a collaboration with Tencent Hunyuan to jointly promote the terminal-side deployment of Tencent Hunyuan Big Model 7B and 3B versions based on the Snapdragon 8 Extreme Edition mobile platform, demonstrating the excellent operational performance achieved by this collaboration. This will help Tencent Hunyuan Big Model provide technical support for a wide range of business scenarios, accelerate product innovation by leveraging terminal-side AI, effectively reduce operating costs, and further expand the application and popularization of generative AI on the terminal side.

The Snapdragon 8 Extreme Edition mobile platform is equipped with a new second-generation custom Qualcomm Oryon CPU and an enhanced Qualcomm Hexagon ^™ NPU, which fully utilizes the advantages of the Qualcomm AI engine to bring more powerful terminal-side generative AI processing capabilities. The powerful AI computing power of the Snapdragon 8 Extreme Edition, combined with the Qualcomm AI software stack and industry-leading tool suites, including the Qualcomm AI Model Efficiency Toolkit (AIMET), provides full-stack optimization capabilities for the model. By using hardware-based INT4 quantization technology, the operating efficiency of Tencent's large mixed model on the terminal side can be greatly improved. The terminal -side inference achieves a latency of 150ms for the first token generation and a decoding rate of more than 30 tokens/second.

Tencent Hunyuan Big Model has provided underlying technical support for more than 700 business scenarios and C-end applications within Tencent, including WeChat input method, Tencent mobile manager, QQ, Tencent video, QQ browser, enterprise WeChat, Tencent conference, etc. By implementing terminal-side deployment for Snapdragon 8 Extreme Edition, it can take advantage of the rich advantages of terminal-side generative AI to better meet a wide range of terminal-side business needs. For example, Tencent Mobile Manager's SMS intelligent recognition function is the first to use Tencent Hunyuan's terminal-side model capabilities. Through massive data combined with deep neural networks and pre-training, the model has a strong semantic understanding ability. By combining contextual information to more accurately understand the intent of SMS, the SMS recall rate has been greatly improved by nearly 200%, and the recognition accuracy has been improved by 20%. Since some SMS involve users' personal sensitive information, the terminal-side AI can also effectively protect the privacy and security of users' personal information while ensuring excellent performance.