Skip to content

README.md

Latest commit

 

History

History
68 lines (51 loc) · 2.16 KB

README.md

File metadata and controls

68 lines (51 loc) · 2.16 KB
Raw

GPU Dashboard

Local web dashboard for checking GPU usage across SSH hosts.

Run

uv run uvicorn gpu_viewer.app:app --host 127.0.0.1 --port 8000

Then open http://127.0.0.1:8000.

Configure Servers

Copy the example config and edit it for your local SSH aliases:

cp config/servers.example.json config/servers.json
{
  "refreshIntervalSeconds": 10,
  "servers": [
    {
      "name": "gpu-node-a",
      "sshHost": "gpu-node-a",
      "teams": ["CL"],
      "allocationLabel": "A100 * 4",
      "allocationSource": "Kakao AI scholar",
      "allocationExpiresAt": "2027-02-28"
    },
    {
      "name": "gpu-node-b",
      "sshHost": "gpu-node-b",
      "gpuAssignments": [
        { "team": "Video", "gpuIndexes": [0, 1], "label": "shared pair" },
        { "team": "Embodied AI", "gpuIndexes": [2, 3], "label": "shared pair" }
      ]
    },
    { "name": "local-workstation", "kind": "local", "group": "workstation" }
  ]
}

Each sshHost should be an alias available to the local ssh command. Use "group": "workstation" for SSH hosts that should appear in the workstation section. Use teams when every GPU on a server belongs to the same team, or gpuAssignments with gpuIndexes when only specific GPU indices belong to a team.

config/servers.json, config/notifications.json, and config/profiles.json are ignored by Git because they can contain private host aliases, GPU UUIDs, PIN hashes, session hashes, and Slack webhook URLs.

To start from notification defaults:

cp config/notifications.example.json config/notifications.json

Profiles

The dashboard supports multiple browser-persistent profiles. Create a profile from the login dialog, add an optional profile photo, protect it with a four-digit numeric PIN, and the browser will keep the session in an HTTP-only cookie. Use the Photo button in the header to update the current profile photo later.

Favorites, Slack webhook settings, and alert rules are stored per profile in config/profiles.json. When the first profile is created, existing config/notifications.json settings are imported into that profile once.

Test

uv run --group dev pytest