Article12 min read
Prefill and Decode Need Different Resources
Prefill processes the prompt. Decode generates one token at a time. Learn when running them on separate workers improves speed and capacity—and when it does not.
AI Infrastructure
The systems that train AI models, run them, schedule GPU work, and track performance.
Start here
Articles
Article14 min read
What Actually Happens When You Send a Prompt to an LLM?
Follow one LLM request from the API gateway to the GPU and back. Learn what prefill, decode, and the KV cache do along the way.
Build it yourself
Guides
Guide15 min read
Build a Local AI Infrastructure Troubleshooting Agent
Build a local AI agent that investigates a broken Kubernetes cluster with Ollama, safe kubectl tools, runbooks, and saved history.