About

How Quant publishes tested guides, measured benchmarks, and reference material for local AI inference and the hardware it runs on, including the Linux systems that drive them. The emphasis is deliberately on what has been run and checked on real machines rather than what a changelog implies might work.

How content works here

The core rule: useful answer first, explanation second.

• Guides lead with the working command or fix, then the reasoning, edge cases, and trade-offs.
• Technical pages carry a tested environment block stating the hardware, OS, kernel, drivers, and tool versions the instructions were verified on — with the date of the last test. If a page is older than the software you're running, assume it needs a double-check.
• Benchmarks record enough configuration to reproduce them: exact build, model file, flags, and sample count. Where possible the raw output files are linked for inspection. Numbers are data, not opinions, and claims are scoped to what was actually tested.
• Commands are intended to be copyable. Every code block on a guide is one that ran, not an illustration.
• Content is updated when the tested environment changes; older findings are dated rather than silently rewritten.

Maintainer

How Quant is maintained by a software developer who tests every hardware and software configuration documented here on real machines before it is published. Results are measured, dated, and reproducible — each benchmark page links the raw data it was written from.

Scope

There are no accounts, comments, or chatbots — a few categories, a handful of well-tested items in each, and full-text search. Depth over volume.