xlite-dev/LeetCUDA ? reverse-engineered prompt

Build me an open source CUDA learning project for beginners that feels like a practical book and a code lab at the same time.

I want clear notes and working examples that teach modern CUDA topics like Tensor Cores, BF16, FP8, FP4, GEMM, and Flash Attention, with enough detail that someone can actually learn from it and run the code. Include a set of benchmarkable kernels, a few advanced examples, and a clean way to browse the lessons from the docs. It should also be easy to build and test from the command line, with a simple quick start for people using a recent NVIDIA GPU.

Please make the repo polished, with readable explanations, diagrams or slides if helpful, and a structure that would make sense for an open source book. If you need to, look up current CUDA docs online so the examples stay up to date.