labs-molt

Provides a research-oriented, agent-first reinforcement-learning training framework supporting fully asynchronous multimodal multi-turn training for large-scale experiments scaled to trillion-parameter mixture-of-experts models.