llm inference · measured
Deep, measured engineering content on how LLM inference actually works.
vLLM internals. KV cache mechanics. GPU kernels. Profiling. Quantization. Written for senior engineers who run models in production — every claim benchmarked, every benchmark runnable.
Latest from the lab
Read it before it hits the front page
One deep post at a time. No fluff, no reposts of the docs — measurements you can rerun and mental models that survive contact with production.
Double opt-in. Unsubscribe anytime. No spam, ever.