dsh-gpu
面向 DeepSeek Harness 的 GPU 感知执行层：gpu_status / gpu_exec / gpu_run_bg 工具、逐步 GPU 上下文注入、自动选择 CUDA_VISIBLE_DEVICES 卡
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add dsh-gpu@0.1.4Gpu 以 dsh-gpu 发布，当前版本为 0.1.4。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
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
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 — one-shot command with a selected card: CUDA_VISIBLE_DEVICES=<freest> is passed through the mounted ctx.shell executor's environment. Auto-select or pin gpuIndex; select count cards for multi-GPU commands.gpu_run_bg — long-running GPU jobs (training, inference servers, benchmarks) register as a gpu job in ctx.jobs: returns a job id immediately, read with job_output, stop with job_kill.time-context pattern), rate-limited to one sample per minute.All execution rides the mounted shell executor. Local host, or any remote execution world (e.g. an SSH provider plugin) — dsh-gpu doesn't know or care where the GPUs are; it queries and launches through the same seam the bash tool uses.
dsh-gpu is an out-of-tree bundle plugin. Install and activate it in a profile with the official plugin command:
dsh plugin --profile <name> add dsh-gpu
The package's bundled cordis.patch.yml registers the plugin automatically. To override its configuration, add an entry with the same id to the profile's cordis.patch.yml:
- insert:
- id: gpu
name: dsh-gpu
config:
stepContext: true
Load order note: place it after your execution-world plugins (e.g. an SSH provider) so the shell seam it queries is the one you intend.
- id: gpu
name: dsh-gpu
config:
stepContext: true # per-step GPU snapshot line (default true)
refreshIntervalMs: 60000 # min spacing between injected snapshots
queryTimeoutMs: 10000 # nvidia-smi timeout
busyMemoryPct: 80 # >= this % memory used => busy
busyUtilPct: 50 # >= this % SM util => busy
nvidia-smi ignores CUDA_VISIBLE_DEVICES — it always reports physical indices. gpu_exec selection still works as intended for CUDA programs; just don't use nvidia-smi output inside gpu_exec to verify the pinning.gpuIndex from a gpu_status read in the same step.gpu_run_bg requires the jobs service in the composition (@deepseek-ai/dsh-jobs + @deepseek-ai/dsh-tool-jobs), the same dependency background bash has.gpu_status reports a clean no-gpu result instead of failing.pnpm install
pnpm typecheck # tsc --noEmit
pnpm test # vitest unit and plugin lifecycle tests
pnpm build # tsdown -> lib/
pnpm check:package # publint + Are the Types Wrong
node tests/live-v100.mjs # optional live probe (edit SSH target first)
Test fixtures are recorded from a live 8× Tesla V100-SXM2-32GB host (including one occupied card) — no mocking of nvidia-smi output formats.
MIT