Guided lab: run a reproducible model baseline

Run the same small CPU model with and without an attention cache. Save its exact version, generated token IDs and step timings so you can check whether a later change improves performance without changing the work.

30 MIN · PREMIUM