五五软件下载

mtop

mtop 是一个本地AI服务器的终端仪表板，支持Ollama、llama.cpp、LM Studio和vLLM，显示加载的模型及其VRAM、GPU信息和每个请求的tok/s。

WindowsMIT身份与来源已核实内容充分核实网盘镜像
查看下载方式

mtop 软件介绍

内容充分核实

以下陈述按信息维度分区展示，每条陈述后紧邻列出支持该陈述的来源。

概览

  • mtop 是一个本地AI服务器的终端仪表板，支持Ollama、llama.cpp、LM Studio和vLLM，显示加载的模型及其VRAM、GPU信息和每个请求的tok/s。
    查看证据（3 条）
    GitHub README.mdOne terminal window for whatever you're running locally (Ollama, llama.cpp, LM Studio, vLLM).
    GitHub README.mdIt shows the loaded models and how much VRAM they're sitting on, the GPU, and the requests going through with their tok/s.
    Winget 清单Description: A terminal dashboard for local AI servers (Ollama, llama.cpp, LM Studio, vLLM).

功能

  • 显示加载的模型及其VRAM、GPU信息和每个请求的tok/s，可以卸载超时的模型，支持多主机监控，提供请求检查器功能，支持比较不同模型的性能，提供会话能源使用统计。
    查看证据（10 条）
    GitHub README.mdOne terminal window for whatever you're running locally (Ollama, llama.cpp, LM Studio, vLLM).
    GitHub README.mdIt shows the loaded models and how much VRAM they're sitting on, the GPU, and the requests going through with their tok/s.
    GitHub README.mdIt'll also kick out models that won't leave.
    GitHub README.mdOllama is supposed to unload idle models on its own and usually does, but every so often `ollama ps` shows something that expired ten minutes ago still parked on 8 gigs. mtop flags those as overdue. `-idle-unload 15m` clears them for you.
    GitHub README.md| `u` | unload the selected model |
    GitHub README.md| `c` | swap recent requests for per-model stats |
    GitHub README.md| `i` | inspector: the last request's prompt, completion, and timing breakdown (needs `-inspect`) |
    GitHub README.mdtok/s is decode speed; TOTAL is wall-clock and includes the model load on the first call.
    GitHub README.mdThe TOK/S line also tracks session energy: watt-hours used and tokens per watt-hour, from whole-GPU power.
    GitHub README.mdIt's an estimate (the whole card, not just inference), so read it as a rough efficiency number, not a meter.

平台

  • 支持macOS、Linux和Windows平台，通过brew、scoop和winget安装，也可以从releases获取二进制文件或使用go安装。
    查看证据（5 条）
    GitHub README.mdmacOS or Linux:
    GitHub README.mdbrew install eladser/tap/mtop
    GitHub README.mdWindows:
    GitHub README.mdscoop bucket add eladser https://github.com/eladser/scoop-bucket
    GitHub README.mdOr grab a binary from [releases](https://github.com/eladser/mtop/releases), or build it with `go install github.com/eladser/mtop@latest`.

安装

  • macOS或Linux使用brew安装，Windows使用scoop安装，也可以从releases获取二进制文件或使用go install github.com/eladser/mtop@latest安装。
    查看证据（5 条）
    GitHub README.mdmacOS or Linux:
    GitHub README.mdbrew install eladser/tap/mtop
    GitHub README.mdWindows:
    GitHub README.mdscoop bucket add eladser https://github.com/eladser/scoop-bucket
    GitHub README.mdOr grab a binary from [releases](https://github.com/eladser/mtop/releases), or build it with `go install github.com/eladser/mtop@latest`.

限制

  • llama.cpp需要使用--metrics和--slots标志才能显示kv-cache数字，AMD GPU需要安装rocm-smi，Apple Silicon需要使用sudo运行才能获取GPU利用率。
    查看证据（4 条）
    GitHub README.mdStart it with `--metrics` (and `--slots`) for the kv-cache numbers.
    GitHub README.mdAMD works if `rocm-smi` is installed.
    GitHub README.mdGPU utilization there comes from `powermetrics`, which needs root, so run mtop with `sudo` on a Mac if you want that number too.
    GitHub README.mdApple Silicon gives you the unified-memory figure on its own.

隐私

  • mtop不会收集任何遥测数据，只与用户自己的服务器通信，代理会拒绝跨域和非回环调用者的请求。
    查看证据（3 条）
    GitHub README.md**Does it phone home?**
    GitHub README.mdNo accounts, no telemetry, it only talks to your own servers.
    GitHub README.mdThe proxy also turns away cross-origin and non-loopback callers, so a browser tab can't reach through it to your ollama.

维护

  • 仓库未归档，支持多主机监控、GPU利用率和内存的sparkline显示、请求检查器等功能。
    查看证据（7 条）
    GitHub Release v1.3.0Watch more than one box, see more per request, and a couple of new numbers.
    GitHub Release v1.3.0- Multi-host: give `-ollama` a comma list and mtop stacks the models and GPUs from each machine, tagged by host.
    GitHub Release v1.3.0- GPU util and memory now draw as sparklines over time, next to the live numbers.
    GitHub Release v1.3.0- Request inspector: run with `-inspect`, press `i`, and you get the last request's prompt, completion, and a load/prompt/decode timing split.
    GitHub Release v1.3.0- Session energy on the TOK/S line: watt-hours used and tokens per watt-hour.
    GitHub Release v1.3.0- `compare -openai <url>` runs the comparison against llama.cpp, LM Studio or vLLM, not just ollama.
    GitHub Release v1.3.0- `-mem-alert` and `-temp-alert` to set the alert thresholds instead of the built-in 93% and 87C.

版本

  • 支持多主机监控、GPU利用率和内存的sparkline显示、请求检查器、会话能源统计等功能。
    查看证据（8 条）
    GitHub Release v1.3.0Watch more than one box, see more per request, and a couple of new numbers.
    GitHub Release v1.3.0- Multi-host: give `-ollama` a comma list and mtop stacks the models and GPUs from each machine, tagged by host.
    GitHub Release v1.3.0- GPU util and memory now draw as sparklines over time, next to the live numbers.
    GitHub Release v1.3.0- Request inspector: run with `-inspect`, press `i`, and you get the last request's prompt, completion, and a load/prompt/decode timing split.
    GitHub Release v1.3.0- Session energy on the TOK/S line: watt-hours used and tokens per watt-hour.
    GitHub Release v1.3.0- `compare -openai <url>` runs the comparison against llama.cpp, LM Studio or vLLM, not just ollama.
    GitHub Release v1.3.0- `-mem-alert` and `-temp-alert` to set the alert thresholds instead of the built-in 93% and 87C.
    Winget 清单PackageVersion: 1.3.0

来源索引

mtop 下载方式说明

进入下载页后，可使用手机扫描二维码打开网盘下载。

授权与下载策略

  • 授权协议：MIT
  • 是否开源：未核实
  • 下载策略：网盘镜像

同类软件推荐

Notepad++网盘镜像VSCodium网盘镜像Microsoft Visual Studio Code官方入口Nginx UI网盘镜像Visual Studio / Code for Command Palette网盘镜像BPMN-RPA Studio上游入口人人译视界上游入口SyncBackSE上游入口