Build 01 · Open
Why does this laptop generate at 20.9 tokens per second?
Predict the decode rate of Qwen3-0.6B on the M3 Pro from memory bandwidth and parameter count BEFORE measuring it. Then measure. Then account for the difference.
Time — two 60-minute blocks
Done when
A state a stranger could checkThe predicted tok/s, the measured tok/s and a stated reason for the gap all exist, and a stranger with the same laptop could reproduce all three.
Why it is harder — First build. Small enough to finish in an evening — a first build that does not finish teaches the opposite of the doctrine.
Where it sits
Shipping it opens 04 Predict the VRAM before renting the GPU
What it reads
Just enough to unblock the next step- Efficiency in LLMs The roofline section only. It is the one idea the prediction needs.
- Computer Architecture L9Only if the bandwidth number itself is unclear. Skip it if the prediction already works.
What it ships
- NotebookA decode-rate number for a machine nobody else is measuring
- BuildThe memory roofline, verified
Teach it back
If it cannot be explained plainly, it was not learned- 1Explain itExplain what you built to a friend who has never seen it. Plain words, no jargon, five sentences at most.
- 2Predicted vs realWhat you expected to happen, what actually happened, and why the two differ.
- 3Where it breaksThe question about it you still cannot answer, and the point where your explanation stops working.
- 4KeepThe one number, rule or picture worth keeping. It becomes a bottle and gets re-tested.
No notes on this build yet.
A note is a Learning Log entry with the tag study-roofline-laptop. It is written in Connor's words, filed as a draft, and appears here once it is published from the admin queue. One note can carry several build tags.