Ollama was randomly hanging during inference. GPU would go idle mid-conversation, requests would timeout.

This section contains lab notebook entries: experiments, observations, and work-in-progress findings.