Ascend Dmi

当用户需要对华为昇腾 NPU 进行硬件层面的管理、测试或诊断时使用此 skill。典型场景: - 查看 NPU 卡的状态、温度、利用率 - 测试内存带宽(h2d/d2h/d2d/p2p) - 跑算力/功耗基准测试(TFLOPS、TOPS) - 诊断 NPU 硬件故障或做健康检查 - 对 NPU 卡做压力测试(aicore、内存) - 复位/恢复卡住或异常的 NPU 卡 典型用户问题(即使不提 ascend-dmi 也应触发): - "帮我测一下这台服务器的算力和带宽" - "看看 192.168.1.50 上 NPU 的温度和利用率" - "跑个 fp16 算力测试,看看 TFLOPS 多少" - "d2d 带宽有没有达标" - "3 号卡好像有问题,帮我诊断一下" - "NPU 卡住了,复位一下" - "对 8 张卡做个压力测试" - "跑个功耗测试,edp 模式" Use this skill whenever the user wants to measure, diagnose, or manage Ascend NPU hardware directly — including checking server NPU performance (算力/带宽/功耗), diagnosing hardware faults, or resetting NPU cards. Trigger even when the user only mentions "服务器" + "算力/带宽/NPU状态" without saying "ascend-dmi". Do NOT use for: writing NPU code, installing frameworks, debugging training issues, or Docker configuration.

ascend-ai-coding ed0f2b3 17 files · 57.9 KB Updated

File contents

ascend-ai-coding/awesome-ascend-skills/tree/main/skills/base/ascend-dmi commit ed0f2b3f2f

Frequently asked questions

npx skillmds@latest add ascend-ai-coding/ascend-dmi