Long-horizon LLM agents are frequently implemented using sparse mixture-of-experts (MoE) models, yet the co-design of agentic behavior and MoE structures rem…
机构:Apple
来源:arXiv 2610.07332 | AI4Papers 论文推荐平台