AI frameworks function the constructing blocks of refined, clever AI fashions, providing a group of libraries and instruments that allow builders to design, prepare and validate advanced fashions effectively.
The Chitu framework, which has been open-sourced since Friday, helps mainstream fashions, together with these from DeepSeek and Meta Platforms’ Llama sequence, in line with the corporate.
When examined with the full-strength model of DeepSeek-R1 utilizing Nvidia’s A800 GPUs, the framework achieved a 315 per cent improve in mannequin inference velocity whereas lowering GPU utilization by 50 per cent in comparison with international open-source frameworks, the corporate mentioned.
The initiative is a part of a broader effort by Chinese language AI corporations to minimize dependence on Nvidia, whose high-performance GPUs are topic to US export controls. Nvidia is banned by Washington from promoting its superior H100 and H800 chips from the Hopper sequence to China-based purchasers.