Models & methods

Fine-tuning

Fine-tuning continues training a pretrained model on your series (or your workspace) so the general prior specializes. It costs more than zero-shot and should earn that cost on a backtest.

Use it when zero-shot is close but systematically off — a retail calendar the pretraining corpus under-represents, a sensor with a house quirk.

Fine-tuning can overfit a short workspace as easily as training from scratch. Keep a holdout. Keep a baseline. 'We fine-tuned' is not an accuracy number.

One pass is usually enough to find out. A second pass that gains a tenth of a WAPE point is a hobby, not a launch.