5 upvotes
New Method Trains Leader-Subagent Teams End-to-End
What is MultiAgent-OPD: End-to-End Training of Hierarchical Multi-Agent Systems with On-Policy Distillation?
It introduces on-policy distillation so a leader and its subagents can be trained together, with an LLM judge providing direct scores to subagents. The technique lets hierarchical agents improve task allocation and execution in a single training loop.
Who it is for
Researchers and engineers building hierarchical multi-agent systems can use OPD to train the whole stack more efficiently.