Instead of one large mixed-RL stage, DeepSeek trains a separate specialist expert per domain.
DeepSeek采用了针对特定领域训练专家的方法，这为模型训练提供了新的视角。
1 Matching Annotations
- Last 7 days
-
nlp.elvissaravia.com nlp.elvissaravia.com
-
Instead of one large mixed-RL stage, DeepSeek trains a separate specialist expert per domain.
DeepSeek采用了针对特定领域训练专家的方法，这为模型训练提供了新的视角。