vLLM 团队实测 agent 写 GPU 内核:Kimi K3 助力达 stock vLLM 1.73 倍
lmoroney · x · 2026-10-20
- vLLM team's new write-up examines whether coding agents can write production GPU kernels: NVIDIA's Benjamin Chislett and collaborators used agent-assisted development to build a fused CuTe DSL decode path on GB300 GPUs
- Results: 842 tokens/second at batch 1 on 16 GPUs with DSpark speculative decoding (synthetic acceptance length 6), 1.73x stock vLLM in the same setup
- The verification is the highlight: an audit found the compiler had hoisted activation reads ahead of a synchronization wait, and long-context evals caught numerical issues that standalone kernel tests with mock inputs missed
Key takeaway: when an agent optimizes your code, make sure your tests actually run the changed path, at the target batch sizes and context lengths.
所属事件:vLLM 团队实测编码 Agent 写 GPU 内核，K3 提速至 1.73 倍(2 条相关)→
「编程与Agent」频道最新
- 微软开源 Scope:量化 AI 编码 Agent 用不好你的 SDK 的原因 — DanWahlin · 2026-10-20
- Claude Motion 动画一键发送 Descript,配音字幕素材一条龙成片 — descript · 2026-10-20
- 作者花两天用 Opus 5.5 打造游戏 demo,3D 资产由 Tripo 生成 — techartist_ · 2026-10-20
- 连 Opus 5.5 都没内化「苦涩的教训」:爱用确定性代码而非提示词 — brandon_galang · 2026-10-20
- 构建 400+ 智能体后总结:15 个多数工程师不知道的 AI Agent 记忆模式 — MaryamMiradi · 2026-10-20
- AI 编码工具 OpenCode 遭黑客入侵，盗用用户 API 密钥消费 — TechCrunch AI · 2026-10-20