notes

writing about language models, CUDA, and systems engineering

2026·systems

vllmini: Building a Minimalist PagedAttention and KV Cache Engine

Developing a clean C++/CUDA library for token generation, covering page tables, virtual memory metaphors for GPU allocation, and block size tradeoffs.

~ ~ ~ ~ ~ ~ ~