Fine-Tuning, and When Not To

Fine-tuning is the most over-reached tool in applied machine learning. This course opens with the honest answer — a better prompt or retrieval solves most of what people want fine-tuning for — and then, for the cases that survive that test, teaches the practice end to end: what training actually changes and what it cannot add, LoRA and QLoRA memory arithmetic you can do on paper, building a dataset and the held-out set that has to come before it, chat templates and loss masking, running a job on a free T4 or a rented card for pennies, catching catastrophic forgetting, measuring whether it helped, and serving the adapter. Written for a mid-range laptop as readily as a rented A100.

1. 1The Honest Answer First7 minA prompt changes in ten seconds; a fine-tune changes in a day. Exhaust the fast loop first.
2. 2What Fine-Tuning Changes, and What It Cannot Add8 minFine-tuning teaches the shape of an answer, not its content; facts belong in the context window.
3. 3Full, LoRA, QLoRA, and the Memory Arithmetic10 minFull fine-tuning costs about 16 bytes per parameter; LoRA charges that only on the 1% you train.
4. 4The Dataset Decides Everything11 minYour model learns your dataset's whole distribution, including the habits you never meant to teach.
5. 5Chat Templates and Loss Masking9 minDecode one batch's unmasked labels before every run; it catches most fine-tuning bugs in five seconds.
6. 6Running a LoRA, and What It Costs10 minGPU time is the cheap part: a 7B LoRA run costs cents, while the dataset and the evaluation cost days.
7. 7Did It Help, and What Did It Break10 minA win rate without an interval and a canary set is not evidence, it is a feeling.
8. 8Serving the Adapter9 minMerging, quantizing and switching engines all change the model; evaluate the exact artifact you serve.