On November 10, Lenovo Group launched the “Wanquan Hybrid Etymology Factory” brand in Zhangjiakou, Hebei. The first parent factory officially landed at the local Beijing-Tianjin-Hebei computing power hub. In the context of the tight supply of AI computing power in China and the explosion in demand for inference side tokens, Lenovo is exporting internally verified full-stack tuning capabilities to the outside world in an attempt to open up the token supply link from lower-level computing power to higher-level applications, and promote commercialization along the industrial agglomeration path. The Wanquan hybrid etymology factory released this time is the middle link in Lenovo's “three-tier hybrid” architecture. Liu Jun, executive vice president of Lenovo Group and president of China, explained that in the three-tier architecture, the bottom layer is Wentian AI infrastructure, covering servers, storage, networks and heterogeneous intelligent computing platforms; the middle layer is responsible for computing power scheduling and word element production; and the upper layer is DynaSky smart services for large government enterprises and Baiying services for small and medium-sized enterprises. According to Yao Zhenyu, vice president of Lenovo Group and general manager of global infrastructure and smart cloud, the “blending” of the hybrid word element factory is reflected in three levels: mixing the deployment of private term pools with public domain elements, the selection and mixing of open source weight models with closed source business models, and the mixing of the computing power of heterogeneous chips. Its core scheduling mechanism is model intelligent routing. According to the request scenario, the system assigns complex tasks to large parameter models and simple interactive distribution to small models, thereby improving overall term production efficiency.

On November 10, Lenovo Group launched the “Wanquan Hybrid Etymology Factory” brand in Zhangjiakou, Hebei. The first parent factory officially landed at the local Beijing-Tianjin-Hebei computing power hub. In the context of the tight supply of AI computing power in China and the explosion in demand for inference side tokens, Lenovo is exporting internally verified full-stack tuning capabilities to the outside world in an attempt to open up the token supply link from lower-level computing power to higher-level applications, and promote commercialization along the industrial agglomeration path. The Wanquan hybrid etymology factory released this time is the middle link in Lenovo's “three-tier hybrid” architecture. Liu Jun, executive vice president of Lenovo Group and president of China, explained that in the three-tier architecture, the bottom layer is Wentian AI infrastructure, covering servers, storage, networks and heterogeneous intelligent computing platforms; the middle layer is responsible for computing power scheduling and word element production; and the upper layer is DynaSky smart services for large government enterprises and Baiying services for small and medium-sized enterprises. According to Yao Zhenyu, vice president of Lenovo Group and general manager of global infrastructure and smart cloud, the “blending” of the hybrid word element factory is reflected in three levels: mixing the deployment of private term pools with public domain elements, the selection and mixing of open source weight models with closed source business models, and the mixing of the computing power of heterogeneous chips. Its core scheduling mechanism is model intelligent routing. According to the request scenario, the system assigns complex tasks to large parameter models and simple interactive distribution to small models, thereby improving overall term production efficiency.