Search papers, labs, and topics across Lattice.
This paper introduces Interactive Training 2, an open-source control plane that allows for real-time adjustments to live model training through a unified interface. By enabling both human and automated controllers to submit requests while maintaining safety through validation at control points, the system enhances the flexibility and oversight of training processes. The authors showcase its effectiveness across five NLP and reinforcement-learning workflows, providing a reusable framework for auditable training practices.
Real-time adjustments to model training are now possible without the need for trainer-specific code, enhancing both flexibility and oversight.
Experiment trackers show how training is progressing, but changing a live run still usually requires trainer-specific code. We present Interactive Training 2, an open-source control plane for steering training through a shared protocol. Training applications declare which settings and actions they expose, humans and automated controllers submit requests through the same interface, and the training loop validates and applies them at safe control points. A customized Aim workspace combines live metrics and controls with a chronological record of requests and outcomes. We demonstrate the system across five NLP and reinforcement-learning workflows. The released code and traces provide a reusable foundation for auditable human- and agent-guided training.
