~/thibault notes

Notes

Write-ups of things I learn while studying and building. Some short, some longer, all in plain language.

2026-11-11Running a 12B model on a 12 GB GPU, and what the whole stack taught meGemma 4 12B at 4-bit on a headless VM with an RTX 3060: why running every layer yourself, from the OS to the model, is the best way to learn how LLM inference really works.