A Guy in Tech

Notes from a homelab that runs too hot. Local LLMs, self-hosted
everything, home automation, and the occasional 3D print. 
Written while the GPU fans spin up.

  • Install llama-swap on Ubuntu (the Easy Way)

    If you run local models with llama.cpp, you already know the annoyance: llama-server serves one model at a time. Want a different model? Kill it, restart with new flags. llama-swap fixes that. It’s a tiny Go proxy that sits in front of your models, reads the model field from each request, and starts (or swaps…


  • Why Hermes Agent Is Eating Your Context Window

    I have been playing with custom built agents for a while, mostly for learning purposes, that work with small local LLMs (typically 7b to 35b). But recently I decided to give a change to a bigger open-source agent that opens a bigger range of possibilities and doesn’t involve me developing every tool I need, so…


  • ROCm vs Vulkan on the RX 7900 XTX

    I benchmarked both llama.cpp backends on a 24GB RX 7900 XTX. The answer isn’t “pick one” — it depends entirely on whether your model is dense or MoE.