A new one Artificial intelligence (AI) Framework developed by teams connected to China Tsinghua University should be able to reduce dependency on dependency Nvidia Chips for AI model inference and mark the latest efforts in the country to improve technological self-sufficiency.
Chitu, a high-performance infection framework for large language models (LLMS), can work on chips in China to the dominance of the graphics processing units (GPUS) of the NVIDIA with the support of certain models such as Deepseek-R1According to a joint explanation of start-up Qingcheng.ai and a team under the direction of the computer science professor Zhai Jidong at Tsinghua University on Friday.
AI frameworks serve as building blocks of highly developed, intelligent AI models and offer a collection of libraries and tools with which developers can efficiently design, train and validate complex models.
According to the company Mainstream models, the Chitu-Framework, which has been open since Friday, supports the Lama series from Deepseek and Meta platforms.
With the tested version of Deepseek-R1 using NVIDIA A800 GPUs, the frame achieved an increase in model infection speed by 315 percent and reduced the use of the GPU by 50 percent compared to foreign open source framework conditions.
05:00
Does the arrival of China's cost-effective deepseek mean the end of the chip dominance of Nvidia?
Does the arrival of China's cost-effective deepseek mean the end of the chip dominance of Nvidia?
The initiative is part of a broader effort of Chinese AI companies to reduce the dependence on Nvidia, the high-performance GPUs of which are subject to US export controls. NVIDIA is prohibited by Washington to sell its Advanced H100 and H800 chips from the Hopper series to China.
