Codebook Agent：用向量量化自动编码器设计多智能体拓扑结构
Great paper on designing multi-agent systems.
多智能体拓扑设计是构建复杂Agent系统的关键瓶颈，这篇论文给出了可复现且高效的自动化设计方案，数据扎实，值得做Agent架构的同学深入研究。
Great paper on designing multi-agent systems.
一篇关于设计多智能体系统的优秀论文。
How many distinct communication topologies does an LLM multi-agent system actually need?
LLM 多智能体系统实际上需要多少种不同的通信拓扑结构？
This works claims that it's about six.
这项研究表明，大约需要六种。
Technical summary:
技术摘要：
Researchers grew the codebook capacity from 8 to 64 and the topologies that survived a reward filter kept collapsing to roughly the same six.
研究人员将码本容量从 8 扩展到 64，而通过奖励过滤器筛选后幸存的拓扑结构最终都收敛到大致相同的六种。
Two further findings undercut the standard formulation. Edge count correlates negatively with measured token consumption at r about -0.4, so sparsifying the agent graph makes inference more expensive.
两项进一步的发现削弱了标准公式的有效性。边数与测得的 token 消耗量呈负相关（r 约为 -0.4），因此稀疏化智能体图会使推理成本更高。
And a message-passing scorer over agent-profile nodes is adjacency-invariant whenever agents share a profile, which is the default configuration in published benchmarks, so it cannot rank candidates at all in that regime.
此外，当智能体共享配置文件时（这是已发布基准测试中的默认配置），基于智能体配置文件节点的传递消息评分器具有邻接不变性，这意味着在该设置下它根本无法对候选者进行排序。
Codebook Agent drops the search entirely. A vector-quantized autoencoder compresses successful topologies into a query-independent 16-entry codebook, a reward-weighted MLP maps the query embedding to a distribution over codes, and an MLP proxy reading the flattened adjacency reranks the top decoded candidates in one batched forward pass.
Codebook Agent 完全放弃了搜索过程。一个向量量化自编码器将成功的拓扑结构压缩为一个与查询无关的 16 项码本；一个奖励加权的 MLP 将查询嵌入映射为码本上的分布；另一个读取展邻接矩阵的 MLP 代理则在一次批量前向传播中对解码出的排名靠前的候选者进行重排。
It emits a topology in 2.4 ms, leads all six benchmarks at 84.6 average against 83.0 for the strongest prior designer, and uses 21.9 to 33.2% fewer LLM tokens.
它在 2.4 毫秒内生成拓扑结构，在全部六个基准测试中平均得分达到 84.6，优于之前最强的设计器的 83.0，并且使用的 LLM token 减少了 21.9% 至 33.2%。
Paper: https://arxiv.org/abs/2609.02264
论文：https://arxiv.org/abs/2609.02264
Chat with Paper: https://academy.dair.ai/papers/codebook-agent-amortized-topology-design-for-llm-multi-agent-systems-2609.02264
与论文对话：https://academy.dair.ai/papers/codebook-agent-amortized-topology-design-for-llm-multi-agent-systems-2609.02264
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力