Evolving Agent Teams：多智能体框架在 GPU 内核优化上超越 FlashInfer
Evolving Agent Teams
一个多智能体框架从零写出了比 FlashInfer 还快 1.2 倍的 CUDA 内核，还附了开源代码，做内核优化的可以试试。
多智能体框架 Evolving Agent Teams 针对现有编程智能体在专家级内核上陷入局部最优的问题，用中央监督者诊断并重组智能体团队，配合包含学习文档和演化历史的持久文件系统积累经验。在相同 token 预算下，该框架持续跑赢主流编程智能体基线。扩大预算后，它从零写出的 CUDA 内核在 FlashInfer-Bench 47 个生产负载中赢下 36 个，平均加速达到 FlashInfer MLA paged decode 内核的 1.2 倍。其他任务上也有具体成绩：AMD MI350X 上 KDA prefill 相比 vLLM 达 3.01 倍，NVIDIA B200 上相比 FlashKDA 达 3.44 倍，Apple M4 Pro 上 MoE 专家卸载服务达到 30.3 tokens/s 解码速度。
Evolving Agent Teams
The rapid progress of software engineering agents drives interest in applying them to open-ended problems such as GPU kernel optimization. However, we find that existing coding agents converge to local optima on real-world, expert-written kernels (e.g., FlashInfer). We introduce Evolving Agent Teams, a multi-agent framework that escapes such local optima. It features (1) a central supervisor that diagnoses each local optimum and restructures a team of agents; and (2) a persistent file system that comprises a learning document, an evolution history, and per-team artifacts, grounding restructuring with accumulated experience. Under a matched token budget, Evolving Agent Teams consistently outperforms strong coding-agent baselines. With an extended token budget, Evolving Agent Teams generates a CUDA kernel that reaches 1.2x mean speedup over the production FlashInfer MLA paged decode kernel from scratch and wins on 36 of 47 production workloads from FlashInfer-Bench (0.86-0.99x on the remaining 11). It also demonstrates superior performance on tasks such as KDA prefill on AMD MI350X (3.01x over vLLM) and NVIDIA B200 (3.44x over FlashKDA), MoE expert-offload serving on Apple silicon (30.3 tokens/s decode for Qwen3.8-Flash-Next in MLX 4-bit on a 48 GB M4 Pro), and RoboDojo robot manipulation (62.1 on held-out seeds vs. 55.3 for the best leaderboard entry). Code is available at https://github.com/caoshiyi/EAT.