Training datasets of the paper: Chain-of-Agents: End-to-End Agent Foundation Models via Multi-Agent Distillation and Agentic RL