Why your local LLM is suddenly 10× slower
VRAM spill, explained
No crash, no error — just a sudden slowdown. What VRAM spill actually is, why it's silent, and how to stop it.
The problems every local-LLM user eventually hits — explained once, properly, instead of re-derived from a forum thread each time. For step-by-step "how do I run this specific model" instructions, see the Models hub instead.
No crash, no error — just a sudden slowdown. What VRAM spill actually is, why it's silent, and how to stop it.
llama.cpp has well over a hundred flags. Here are the handful that separate barely-working from genuinely fast.
How a 120B-parameter model runs on hardware that can't hold 120B parameters — the mechanics, explained.
Why community forks exist, which ones are worth knowing, and how to run one without compiling anything yourself.
These guides cover the concepts. For step-by-step setup, see the Models hub (pick a model) or What can I run? (pick your hardware).