Skip to content

Launching a Notebook

Creating a notebook server

  1. Open https://kubeflow.pais.auckland.ac.nz
  2. Select your namespace (e.g., rg-compsci) from the top dropdown
  3. Click Notebooks in the left sidebar → + New Notebook
  4. Configure:
    • Name — a unique name for this server (e.g., nlp-experiments)
    • Image — select jupyter-pytorch-cuda for GPU-accelerated PyTorch work
    • CPU / RAM — start with 2 CPU / 8 Gi; increase if needed
    • GPU — set to 1 if you need GPU access (uses one time-slice of the L40S)
    • Workspace Volume — a PVC that persists across restarts (10 Gi default)
    • Data Volumes — attach your research group's shared VAST storage
  5. Click Launch

The server takes ~2 minutes to start on first launch (image pull). Subsequent starts are faster.

Connecting to the notebook

Once the server shows Running, click Connect to open JupyterLab in your browser.

Environment variables in notebooks

The PAIS API endpoint and your API key are available as environment variables inside notebooks:

import os
api_base = os.environ.get("PAIS_API_BASE", "https://api.pais.auckland.ac.nz/v1")
api_key = os.environ.get("PAIS_API_KEY", "")

Tip

If PAIS_API_KEY is not set automatically, add it to the notebook server's environment variables when creating the server (under Advanced Options → Environment Variables).

Stopping and deleting servers

  • Stop (from the Notebooks list) — deallocates GPU/CPU, keeps the workspace PVC and its data
  • Delete — removes the server and its workspace PVC; data is lost unless saved to shared storage

Always save important results to your group's VAST storage before deleting a server.

Resource limits

Notebook servers are subject to the research group's ResourceQuota. If your server fails to start with a "quota exceeded" error, either reduce the requested resources or coordinate with your group to free up capacity.