dsh-gpu

v0.1.4 · 3 天前发布

GPU-aware execution layer for DeepSeek Harness: gpu_status / gpu_exec / gpu_run_bg tools, per-step GPU context injection, automatic CUDA_VISIBLE_DEVICES card selection

deepseek-harnessdsh-plugingpunvidiacudanvidia-smi

README

dsh-gpu

GPU-aware execution layer for DeepSeek Harness (dsh). Out-of-tree plugin; no harness patches required.

Agents get three tools — gpu_status, gpu_exec, gpu_run_bg — plus an optional per-step GPU context line. Cards are selected automatically (freest first) with CUDA_VISIBLE_DEVICES set in the command environment; pin a card explicitly when you care.


8 GPU(s), free: [0,1,2,3,4,5,6,7]
GPU0 Tesla V100-SXM2-32GB: 4264/32768MiB 0%util 40C
...
[gpus 1 — GPU 1 (auto: freest 1)] exit 0

How it works

gpu_status — one query, every device: memory used/total, SM utilization, temperature, and a free/busy verdict. A device is busy at or above 80% memory used or 50% utilization (both configurable).

*gpu_exec

查看完整 README →

版本兼容矩阵

DSH 版本	状态	说明
`dsh@0.1.0-rc.6`	✅ 兼容	cordis patch ✅
`dsh@0.1.0`	✅ 兼容	标准 patch ✅
`dsh@0.2.0`	⚠️ 待验证	待实测（API 可能在变）
`dsh@0.3.0+`	❓ 未知	尚未发布

💡 兼容矩阵每周末自动跑实测，欢迎 PR 修正

元数据

License: MIT
npm: dsh-gpu
Repo: git+https://github.com/zytsyj/dsh-gpu.git
Engines: {"node":"^22.19.0 || >=24"}
Weekly DL: 0
Monthly DL: 0
dsh.bundle: {"bundle":{"patch":"./cordis.patch.yml"}}