China Telecom Unveils AI Model with 1 Trillion Parameters, Boosting China’s Self-Sufficiency in Domestic LLM Training

China Telecom has reported that it trained an AI model featuring 1 trillion parameters using Chinese-made chips. Although the government-controlled telecommunication company didn't disclose the specific chips utilized for its complex model, it is known to collaborate with Huawei.

This accomplishment signifies that China has genuinely achieved complete independence in their local LLM training. It also flags the commencement of a fresh stage for China's innovation and self-dependency in LLMs, the tech that powers OpenAI's ChatGPT, according to a statement released on WeChat by the AI institute.

China Telecom announced that an undisclosed model possesses 1 trillion parameters – a term used in machine learning to denote the variables in an AI system during its training phase. The complexity and efficiency of an AI model are heavily influenced by the quantity of parameters used in the training procedure. The company stated that TeleChat2t-115B comprises more than 100 billion parameters.