About
Token Budget is a personal notebook about building things with language models — written by an infrastructure engineer who spends a lot of evenings running models on modest hardware and taking them apart.
The posts here are the long version of notes I would otherwise keep in a text file: how much memory a model really needs, why a retrieval pipeline returned the wrong passage, how to tell a real improvement from noise. The goal is to be practical and specific, and to show the arithmetic.
What you won't find here
- Predictions about AGI timelines.
- "Top 10 prompts" lists.
- Analytics, cookies, newsletter pop-ups or third-party scripts. This site is a handful of static files; your browser doesn't talk to anyone but this server.
Contact
Corrections are always welcome — especially the ones that come with numbers. Write to hello@tokenbudget.duckdns.org. New posts appear in the RSS feed.