We propose a highly efficient distillation algorithm for flow matching models that requires minimal computation and enables few-shot training.