/finetuning-method-selection

PurposeA router skill for the fine-tuning lifecycle that decides whether to fine-tune and selects the right method (SFT, DPO/ORPO/KTO, GRPO/RLVR, CPT) and base-model size class. It uses off-ramp checks, a decision tree by data shape, and catalog references to route efficiently. Value: prevents wasted training runs by matching task to cheapest effective approach before any model work begins.