Cube Studio supports SFT fine-tuning, reward modeling, and reinforcement learning training for large models such as DeepSeek.