■ Training

Fine-tune on a single GPU or scale to multi-node runs without reserving capacity.