Writing
Posts grouped into series. Each series goes deep on one topic.
Benchmarks with RTX 3080
What I wanted from a 10 GB RTX 3080, the tests I wrote to find out, and how I tuned engines and models until it fit. Test code and cases are published.
- 1The chip, and what I want from it5 min
- 2The tests I wrote6 min
- 3vLLM first, and why it did not stay5 min
- 4Fitting a real model in 10 GB4 min
- 5Raw speed against the ceilings6 min
- 6Engines head to head6 min
- 7Does a bigger model pay off?9 min
- 8The 35B with parallel slots and image input5 min
- 9A faster challenger: Gyro-S6 min
WTF is Agent?
How an autonomous agent actually works, one component at a time, from my own build. The anatomy, the design decisions, what I deleted, what I measured, and a survey of what already exists.
- 1The anatomy of an agent7 min
- 2Memory: remember, cite, forget6 min
- 3The toolbox: one door to every tool6 min
- 4The harness: a loop that survives6 min
- 5Context and skills: what the model reads6 min
- 6Runtime and sandbox: where agent code runs6 min
- 7Waking, asking, stopping7 min
- 8Knowing it works: evals13 min
- 9What I deleted5 min
- 10AI is a label, not a thing3 min
- 11Machine learning is fitting a curve3 min
- 12Neural networks are lots of tiny dials2 min
- 13Language models are autocomplete at scale3 min