面壁智能OpenBMB感谢@aijoey用MiniCPM5-1B构建后端智能体集群。128个并发智能体在DGX Spark上运行，通过vLLM连续批处理提供服务，每个智能体独立处理发票审核、退款路由、合规检查等8种业务队列。系统在1.48秒内跨智能体流式传输6604个chunks。该案例表明，1B模型的价值在于同时做出大量有用业务决策——用一群小型廉价worker并行清理队列。
Huge thanks to @aijoey for building back-office agent swarm with MiniCPM5-1B 👏
This is a fantastic real-world case of scaling small models into production-grade systems——moving beyond “model capability” into “practical multi-agent systems at scale”.
We’re especially impressed by the technical setup:
🔷128 concurrent agents on DGX Spark
🔷vLLM continuous batching for serving efficiency
🔷6,604 chunks streamed across agents in just 1.48s
🔷Clear demonstration of how a 1B model can power high-throughput, multi-agent workflows in parallel
Really impressive work on the back-office swarm setup and the 128-agent parallelization. Excited to see what else you build with MiniCPM in the future 🚀
Built a back-office agent swarm demo with MiniCPM5-1B. 128 concurrent agents on DGX Spark, served through vLLM continuous batching. Each agent handled a different business ops queue: invoice review refund routing compliance checks HR/payroll issues legal triage IT security alerts sales ops logistics recovery MiniCPM5-1B streamed 6,604 chunks across 128 agents in 1.48s. The takeaway: The value of 1B models is not just “can it answer well?” It is: “How many useful business decisions can it make at once?” Not one giant agent doing everything slowly. A swarm of tiny, cheap workers clearing practical queues in parallel. cc: @OpenBMB @NVIDIAAI在 X 查看被引用的帖子
来源：OpenBMB · x.com