Inspiration from llama.cpp: Implementing a 4.3B Tokens/s Stateless Data Engine

[Anh-Khoa-PC]

Hi Georgi (@ggerganov),

I'm Khoa, a 15-year-old developer from Vietnam. Inspired by your work on making AI accessible on consumer hardware, I've developed Vantage V-AI, a stateless data engine that achieves 4.3 Billion Tokens/s on a standard 12-thread CPU with minimal RAM footprint (< 2MB).

Like your projects, Vantage focuses on CPU optimization and memory efficiency using LZ4-stream and zero-copy batching. I’d love to get your thoughts on the architecture or any potential integration with high-performance C++ workflows.

Repo: [](https://github.com/Anh-Khoa-PC/VANTAGE-V-AI)

Keep up the amazing work with llama.cpp!