``` ### Distilled types DeepSeek team has shown which the reasoning patterns of more substantial models can be distilled into smaller sized versions, resulting in much better overall performance compared to the reasoning designs found through RL on small models. Notably, it's the initial open