LLM agents execute tasks through repeated model calls over evolving histories of reasoning, actions, and tool feedback. Aggregate speculative-decoding metric…
机构:华为
来源:arXiv 2610.41177 | AI4Papers 论文推荐平台