DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

System Monitor Xg — DSH Plugin for DeepSeek Harness
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in
← Plugins

dsh-system-monitor-xg

System Monitor Xg

DSH system load monitoring plugin: Real-time display of CPU/memory/GPU compute and VRAM bandwidth utilization, VRAM usage, power consumption, and temperature; automatic diagnosis of inference bottlenecks (compute vs. bandwidth vs. power vs. thermal vs. VRAM capacity); Web bottom bar + agent tools +

The plugin will be installed here. Keep web if you are unsure.

npx -y @deepseek-ai/dsh plugin --profile web add dsh-system-monitor-xg@0.1.4
READMECompatibilityVersions

Description

DSH system load monitoring plugin: Real-time display of CPU/memory/GPU compute and VRAM bandwidth utilization, VRAM usage, power consumption, and temperature; automatic diagnosis of inference bottlenecks (compute vs. bandwidth vs. power vs. thermal vs. VRAM capacity); Web bottom bar + agent tools + prefill/decode stage comparison + load JSONL persistence.

Compatibility and provenance

System Monitor Xg is published as dsh-system-monitor-xg and currently resolves to version 0.1.4. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
web
Release source
npm
Registry updated
9/20/2026

Versions

0.1.4stable
8/20/2026
0.1.3stable
8/20/2026
0.1.2stable
8/20/2026
Show 1 more versionCollapse versions
0.1.1stable
8/20/2026

Related plugins

Loading related plugins…

Latest
0.1.4
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
146.6 kB
Files
37
Surface
web
License
MIT
Source
npm
GitHub
★ 2
Weekly downloads
44
Last push
9/1/2026
View source ↗Project homepage ↗
README badge

Click the badge to copy Markdown for your README.

Do you maintain this Plugin?Claim benefit · Priority security scan

Verify the GitHub repository declared in package.json to manage this listing. After you claim it, Hub will prioritize a security scan of the current version and publish the result when it passes.

Claim this Plugin →
Report an issue

Related plugins

More verified plugins in ui-customization.

Web App@deepseek-ai/dsh-web-appThe dsh browser-surface bundle: the web patch layer over dsh-base plus the runtime glue plugin (frontend dist serving, web-surface prompt, bash runtime variables, URL line)Experimental Agent Team Web Profile@deepseek-ai/dsh-experimental-agent-team-web-profileExperimental Web profile layer for Agent Teams Remote and UI pluginsRemote Web Ui@linxin666/dsh-remote-web-uiScan-to-pair remote access for the dsh web GUI that shares one official interface: a QR beside the settings button pairs phones and PCs into the same Web GUI (a portrait-touch adaptation layer for phones, full desktop on PCs) through one-time tokens and rClient Ui Task Board@linxin666/dsh-client-ui-task-boardHost-authoritative task board for the DSH Web GUI with real session execution, Host cron scheduling, and optional cross-platform idle-sleep protection; mounted without DSH source changes.

README

dsh-system-monitor-xg

DSH(DeepSeek Harness)宿主机负载监控插件:CPU / 内存 / GPU 算力与显存带宽 利用率、显存、功耗、温度实时显示,并自动诊断推理瓶颈——「GPU 占用高 ≠ 算力打满」:decode 阶段 SM 大量时间在等显存返回数据,瓶颈常常是显存带宽 而不是算力。

解决什么问题

nvidia-smi 有两个正交的利用率数字:

指标含义
utilization.gpuSM 算力利用率(计算单元活跃占比)
utilization.memory显存带宽控制器利用率(≈ 带宽占用,注意不是显存占用!)

两者组合即可回答「GPU 很忙,忙在算力还是带宽」:

现象诊断
SM 高 + 带宽高 + 功耗低带宽受限(decode 典型:算力等显存)
SM 高 + 带宽低算力受限(prefill 典型:计算饱和)
功耗 ≈ 上限功耗墙
显存占用 > 95%显存容量瓶颈(上下文受限)
高温 + SM 时钟降频热限制

特性

  • Web 底栏实时显示:CPU%、内存%、每张 GPU 的 SM 算力 / 显存带宽 / 显存 / 功耗 / 温度,叠加瓶颈徽标(算力=蓝、带宽=琥珀、功耗=橙、热/显存=红)
  • 生成阶段对比:点击底栏展开最近 10 次生成的 prefill vs decode 两阶段负载统计——prefill 算力密集、decode 带宽密集,两阶段对比是 「占用高但算力不满」的量化证据
  • agent 工具 system_metrics:会话内随时查询负载快照、瓶颈诊断、 生成阶段对比、显存占用进程(评测报告可直接引用)
  • 负载 JSONL 落盘:~/.dsh/dsh-system-monitor/metrics-<日期>.jsonl (每秒采样)+ generations-<日期>.jsonl(每次生成的阶段汇总)
  • 零运行时依赖:CPU/内存用 Node 内置 API,GPU 解析 nvidia-smi CSV ——无需 pip、无需原生模块、无需管理员权限
  • 多 GPU 支持(取最活跃 GPU 做瓶颈诊断)

安装(给使用者)

# 1. 安装插件(从 npm,无需 clone 仓库)
dsh plugin --profile web add dsh-system-monitor-xg

# 2. 重启 DSH Web(生产模式无热装机制;重启方式与你平时启动 dsh 一致)
#    重启后打开任意会话页,输入框下方(内置统计行右侧)出现负载条:
#    CPU 12% 内存 34% │ GPU0 SM 45% 带宽 88% 显存 12/24G 160W 66℃ ●带宽受限

# 3. 验证
#    - 底栏每秒刷新;跑一次本地推理可见瓶颈徽标变化(带宽受限/算力受限…)
#    - 点击底栏展开最近生成的 prefill/decode 阶段对比
#    - 会话里让 agent 调用:system_metrics(查询负载/瓶颈/阶段统计)
#    - 负载数据落盘 ~/.dsh/dsh-system-monitor/(metrics-<日期>.jsonl)

# 4. 升级
dsh plugin --profile web update dsh-system-monitor-xg

# 5. 卸载
dsh plugin --profile web remove dsh-system-monitor-xg

环境要求:NVIDIA GPU + 驱动自带 nvidia-smi(Windows 在 System32,Linux 在 /usr/bin);无 NVIDIA GPU 时插件正常加载,GPU 段显示「不可用」,CPU/内存 监控不受影响。开发模式(想改代码调试)用 dsh plugin --profile web add <本仓库路径>。

使用

  • 底栏每 1 秒刷新一次,直接可读: CPU 12% 内存 34% │ GPU0 SM 45% 带宽 88% 显存 12.4/24G 160W 66℃ ●带宽受限
  • 悬停查看细节(单核分布 / 时钟 / 功耗上限);点击展开生成阶段对比表
  • Agent 会话中直接调用工具(例):
    使用 system_metrics 工具查询当前负载和瓶颈,history=5, generations=5
    

REST API(host 侧,供脚本/底栏轮询)

端点说明
GET /system-monitor/api/snapshot?light=1最新采样(light 只回 now,底栏轮询用)
GET /system-monitor/api/snapshot最新采样 + 最近 60 条历史 + 最近 10 次生成
GET /system-monitor/api/generations?limit=20生成阶段汇总
GET /system-monitor/api/history?window=120采样历史窗口
GET /system-monitor/api/processes显存占用进程(Windows 上可能无权限读显存值)

配置(cordis.patch.yml insert 的 config)

- insert:
    - id: dsh-system-monitor-xg
      name: 'dsh-system-monitor-xg'
      config:
        intervalMs: 1000    # 采样间隔(250–10000ms)
        historySize: 3600   # 内存环形缓冲条数
        persist: true       # JSONL 落盘开关

开发

powershell -NoProfile -ExecutionPolicy Bypass -File scripts/build.ps1
# 构建产物:lib/index.js(host)+ lib/client.js(浏览器 bundle)+ lib/types/
  • 构建脚本自动 junction 类型依赖到 D:\deepseek-harness($env:DSH_CHECKOUT 可覆盖),用 checkout 的 tsc / tsdown,本仓库无需 npm install
  • 安装为 link: 方式时,改代码后重新构建即生效(重启 DSH Web 后)
  • 结构:src/ host(采样 cpu.ts / gpu.ts、诊断 bottleneck.ts、 阶段跟踪 sampler.ts、落盘 ledger.ts、入口 index.ts)、 src/client/ 浏览器端(SystemBar.tsx 底栏、fetch.ts 同步 XHR 轮询)

关键技术决策

  • 数据通道:底栏 1s 轮询 host REST API(同步 XHR——DSH web 环境会随机 stall 异步 fetch,社区 dsh-status-bar 踩坑后的结论)
  • 阶段边界:request/header(prefill 开始)→ 首个 assistant/chunk (decode 开始)→ assistant/message(结束)
  • 瓶颈阈值:bottleneck.ts 中纯函数,判据顺序 idle → 显存容量 → 热限制 → 功耗墙 → 带宽 → 算力 → 均衡

已知限制

  • Windows 下 nvidia-smi --query-compute-apps 读进程显存需要管理员权限, 无权限时 usedMb 为 0(进程名仍可见)
  • 阶段统计依赖采样点落在 prefill/decode 窗口内;极短生成(< 采样间隔)可能 无采样点
  • 底栏挂在会话页(composer.dock),无会话时不显示(与内置统计行一致)

License

MIT