Typed decisions on your own GPUs
How SGLang's /v1/decisions turns a chat model into a classifier, how to calibrate and test its probabilities, and when to self-host.
Self-hosted LLMs
Every post on this topic, newest first.
Updated 3 OCT 2026
More users per GPU: a playbook for self-hosted LLM serving
What to measure first, then the levers that raise LLM serving throughput on your own GPUs, in the order we pull them and what each one trades away.