The current Sino-US tensions have led to widespread embargoes on AI chips/technologies, making it imperative to establish a domestic computing-powered AI development and application ecosystem.
The VT-Transformer is a AI computing framework, it integrated C++ open-source computing framework addresses foundational AI infrastructure needs with usability, accessibility, and efficiency—serving as an alternative to NVIDIA's Python/CUDA stack. It reduces training costs, enhances inference performance, and enables mainstream open-source models to adapt to China's low-resource computing hardware.
Industrial digitalization requires precise computation of the physical world. Current visual models demand continuous scenario data iteration, while large models suffer from hallucinations that produce erroneous outputs – both significantly hindering AI deployment in industrial settings.
The MaM Hybrid Model Inference Engine integrates diverse models including compact visual models, visual/language/multimodal large models, expert mechanistic models, and equipment digital twins. Through intelligent model orchestration and comprehensive application management, it delivers high-accuracy physical world computations.
