谷歌新论文：用Agent理解瓶颈，将LLM基础设施优化从暴力搜索变为精准定位
New Google paper shows LLM infrastructure optimization does not have to be a gia…
做LLM基础设施优化的同学必看，这篇把优化从暴力搜索变成理解瓶颈后的精准搜索，给出了可复现的Agent架构和验证结果，值得精读并借鉴到自己的调优流程里。
New Google paper shows LLM infrastructure optimization does not have to be a giant brute-force search if an agent can understand why the system is slow.
谷歌新论文表明，如果智能体能理解系统为何缓慢，LLM基础设施优化就不必进行大规模暴力搜索。
The key shift is from “search everything” to “understand the bottleneck, then search the tiny part that matters.”
关键转变是从“搜索一切”到“理解瓶颈，然后搜索关键的小部分”。
Instead of asking, “Which of these 100 configurations is fastest?”, this work asks, “What is actually causing the slowdown?” That one change dramatically shrinks the search.
这项工作不是问“这100种配置中哪种最快？”，而是问“究竟是什么导致了减速？”这一改变极大地缩小了搜索范围。
Its Analyzer Agent reads profiler traces and classifies the bottleneck as compute, memory, or communication, then a Proposal Agent retrieves engineering knowledge and generates three targeted ways to map data, model, and sequence parallelism across TPU chips.
其分析器智能体读取性能分析跟踪，并将瓶颈分类为计算、内存或通信，然后提议智能体检索工程知识，生成三种针对性的方法，在TPU芯片间映射数据、模型和序列并行。
Across eight production workloads spanning dense and MoE models, pre-training, post-training, serving, and TPU systems from 2 to 2,048 chips, the human-validated production configuration appeared in PROMPTS' first batch every time.
在涵盖密集和MoE模型、预训练、后训练、服务以及从2到2048个芯片的TPU系统的八个生产工作负载中，经人工验证的生产配置每次都出现在PROMPTS的第一批中。
In seven of eight cases, it was also the first configuration tested, and the agent's top-ranked proposal matched what engineers ultimately adopted in 87.5% of cases.
在八种情况中的七种里，它也是第一个被测试的配置，并且智能体排名最高的提议在87.5%的情况下与工程师最终采用的方案一致。
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