infrasys-ai/aisystem

Infrasys-AI/AISystem is an open-source educational course focused on the full software and hardware stack that powers AI systems. The course is primarily in Chinese and delivered through slides, Jupyter notebooks, and video lectures. It targets senior undergraduates, graduate students, and AI infrastructure practitioners who want a deep understanding of how AI systems are built. The course covers five main areas. The first is an overview of the complete AI system stack, algorithms, frameworks, and hardware architecture. The second dives into AI chip design, covering CPUs, GPUs, and specialized AI processors, and how chip design must account for AI algorithms and frameworks. The third covers AI compilers, tools that translate high-level model descriptions into efficient machine code. The fourth covers AI inference systems (running a trained model to produce predictions), including model compression techniques like quantization and pruning that make models smaller and faster. The fifth covers AI framework technologies such as automatic differentiation (software that computes the math gradients needed for training), computational graphs, and distributed training. Someone would use this resource when learning how AI works not just at the algorithm level, but at the systems level, understanding how hardware, compilers, frameworks, and inference engines connect. Slides are on GitHub; video lectures are on external video platforms.