arXiv:2609.06702 · Agent 智能体
PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents
PARSER：长上下文 LLM Agent 的并行读取与深度推理
PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents
- 类型：arxiv
- 标识：2609.06702
- 链接：https://arxiv.org/abs/2609.06702
- 主分类：agent
- 形态：method
- TLDR：Sequential memory agents process long documents by reading chunks one after another while maintaining a compact memory state, coupling document traversal to reasoning depth. This coupling introduces sensitivity to evidence placement and ties inference latency linearly to document length. We introduce PARSER, which decouples reading from reasoning. A bank of lightweight subagents each bound to a single chunk read the entire document in parallel, while a lead agent reasons in depth through iterative scatter--gather rounds: at each round it broadcasts a query to all subagents, aggregates the retu
- 待LLM分类：否
- 标题中文：PARSER：长上下文 LLM Agent 的并行读取与深度推理
- TLDR中文：顺序记忆型 Agent 通过逐块顺序读取长文档来维持紧凑的内存状态，将文档遍历与推理深度耦合。这种耦合导致对证据位置敏感，并将推理延迟与文档长度线性关联。我们提出 PARSER，将读取与推理解耦：一组轻量子 Agent 各自绑定单个块，并行读取完整文档；主 Agent 通过迭代的 scatter-gather 轮次进行深度推理：每轮向所有子 Agent 广播查询，汇总返回……
- 来源文件：
- /inbox/tom/_candidates/2026-09-11-agent-rag-longcontext-candidates.json