Skip to content

Creating a notebook

Start a development environment directly in the TUKE Kubernetes cluster.

ml.cloud.tuke.sk

  • No request needed

    Your own namespace is created automatically on first sign-in.

  • Three environments

    JupyterLab, Visual Studio Code or RStudio.

  • Persistent data

    Files stay on disk even after the notebook is stopped.

  • TUKE ID

    The same sign-in as for other services.

Signing in

Open ml.cloud.tuke.sk and sign in with your TUKE ID. Your personal namespace is created automatically and shown in the top bar next to the package icon.

Kubeflow home screen

The main menu on the left:

Item What it does
Home Overview and quick shortcuts
Notebooks Creating and managing notebook servers
TensorBoards Visualising training logs
Volumes Storage volumes and the file browser
Pipelines Pipelines, experiments, runs and artifacts
Manage Contributors Sharing the namespace with colleagues

The namespace is your space

All notebooks, volumes and pipelines belong to your namespace. Other users cannot see into it unless you grant them access via Manage Contributors.

Notebook list

The Notebooks section lists all your notebook servers.

List of notebooks

The table shows status, name, environment type, creation date, last activity, the container image in use and the allocated resources.

Element Meaning
Green marker The notebook is running, CONNECT takes you into it
Grey marker The notebook is stopped, start it with
Square Stops a running notebook
Bin Permanently removes the notebook

A stopped notebook consumes no resources

Data on the workspace volume is preserved, so you can stop a notebook and continue your work later.

Creating a new notebook

Click + New Notebook in the top right.

Name and environment

New notebook: environment and resources

Fill in Name and choose an environment:

  • JupyterLab

    Interactive notebooks for prototyping and data analysis.

  • Visual Studio Code

    A full editor for developing and debugging applications.

  • RStudio

    An environment for statistics and the R language.

Below the environment picker is a dropdown with the container image. The default is Custom Notebook, where you can enter your own image.

CPU and memory

Under CPU / RAM enter the minimum number of cores and minimum memory in Gi. This is the lower bound the notebook is guaranteed to receive.

Field Default
Minimum CPU 0.25
Minimum Memory Gi 1

GPUs are not available

Leave Number of GPUs set to None. ML TUKE runs on CPU only.

Storage volumes

New notebook: volumes and configurations

The Workspace Volume is mounted as your home directory. By default a new volume named after the notebook is created.

Data Volumes are optional additional volumes:

  • + Add new volume creates a new one
  • + Attach existing volume mounts one you already have

Configurations

Under Configurations there is an option called Globálny zdieľaný dataset. Ticking it mounts the folder shared by all users of the platform into your notebook.

The shared folder has its own rules

All signed-in users can see and modify its contents, and it is wiped completely once a year. Read the terms before using it.

Global shared dataset

Additional settings

Setting When you need it
Affinity / Tolerations Targeting specific cluster nodes, normally leave at None
Enable Shared Memory Turn on for PyTorch dataloaders, otherwise they may fail

Create the notebook with LAUNCH. Startup takes a few dozen seconds while the container image is pulled.

Working in the environment

Once the notebook reaches the Running state, open it with CONNECT.

JupyterLab environment

In JupyterLab you can create notebooks via New → Python 3, upload files with Upload and organise projects into folders. Changes are saved continuously onto the workspace volume.

How to check the resources actually allocated

The values in the form are minimums. You can read the container's real limits straight from cgroup:

from pathlib import Path

def read(p):
    try:
        return Path(p).read_text().strip()
    except FileNotFoundError:
        return None

print("memory:", read("/sys/fs/cgroup/memory.max"))
print("cpu:   ", read("/sys/fs/cgroup/cpu.max"))
print("cores: ", read("/sys/fs/cgroup/cpuset.cpus"))

TensorBoards

The TensorBoards section is used to visualise training logs.

TensorBoards

Create a new instance with + New TensorBoard, where you enter the path to the logs (Logpath) on one of your volumes.

Recommendations

Best practices

  • Stop notebooks you are not using, they consume compute resources
  • Data stays stored even after the notebook is stopped
  • Clean up regularly — unneeded files, datasets and logs
  • Use Git for version control of your code

Next steps

  • Storage and files

    Managing volumes and browsing files without a running notebook.

    Guide

  • Global shared dataset

    The shared folder for lab datasets.

    Rules

  • Custom Docker image

    An environment with specific libraries.

    Guide