CourseTransformers: From Attention to Trained, Served and Interpreted Models · Module 12: Adaptation and Alignment · part 93 of 111
Part 93 · Module 12: Adaptation and Alignment

Lab: Package the card model and Maya's check

87 min read·13 Oct 2026
Module 12 · Lab · About 70 minutes · Needs: lessons 1 to 5 of this module, and the Module 11 lab

What you are building

TINYFORMER-LAB AFTER MODULE 12 adapt/cards.py: the task, the scorer, Maya's bar sft.py replay recipe score.py end checked lora.py, prefs.py kept, tested m12_adapt run one seed, 2,400 steps evals/log_cards.py exit 0 only on a pass v0.11.0 package unchanged; generate.py's cache reused test_m12_adapt: 8 tests git tag v0.12.0
The package (blue) holds the task and the recipe; one run (amber) feeds Maya's check (green); earlier files (grey) stay as they were.

Sol 446 at Aurora Station. Five lessons answered Maya Lindqvist's question in notebook cells, and the cells disagree. Lesson 2's fine-tunes wrote no right card. Lesson 3's adapter wrote "Omar Hinters". Lesson 4's stand-in wrote 71.5% exact cards and forgot the corpus. Lesson 5's recipe wrote 42.5% to 62.0% and forgot nothing.

Today the lab settles one account of those results and saves the module in tinyformer-lab: tinyformer/adapt/ with the task, the recipe, LoRA and the preference code; evals/log_cards.py, Maya's bar as a command; eight tests; one experiment. The tag is v0.12.0.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.

Buy this course · ₹399See what is inside