The Tencent WeChat team announced on X the WeLM large language model developed by the WeChat team, with resource efficiency as its core focus.
According to the official poster, the two currently confirmed specifications in the WeLM large language model series are:
WeLM-80B: 80 billion total parameters, 3 billion active parameters
WeLM-617B: 617 billion total parameters, 23 billion active parameters
The WeLM-80B model has been deployed in WeChat's native AI assistant, Xiaowei, supporting the following features:
Chat and search
Use of WeChat's native features
Access to mini program services
IT Home learned that the WeLM-671B model is under development. It is a mixture-of-experts (MoE) model that significantly improves general understanding and logical reasoning while maintaining a moderate number of active parameters, targeting complex tasks within the WeChat ecosystem:
Intelligent mini program development
Xiaowei tool generation
