Lab: The fine-tuning trial, from plan to decision

Module 9 · Lab · About 60 minutes · Needs: lessons 1 to 4 of this module, and the Module 1 lab (or the cells in Step 1)

What you are building

It is Friday 31 July 2026, the end of week 9. Each lesson answered part of Tomasz's question in its own notebook: a frozen plan and a bar (Lesson 1), a clean training file (Lesson 2), a LoRA run scored once (Lesson 3) and a verified reply adapter (Lesson 4). This lab puts the trial into the repository as finetune/ and ends with one written decision for Maya. The running example is T-1025, Ruth Pell's two charges of 144 USD, a test ticket that must never reach training. Twelve calls go to the hosted model, in Step 3; everything else runs SmolLM2-135M-Instruct on your CPU, with under ten minutes of training.