Ten fine-tuning mistakes I see students make (and made myself)

The mistakes that turn a fine-tuning project into a shrug: no baseline, dirty data, the wrong metric, training on eval data, chasing loss, and skippin