How to use each menu

Notebooks

Notebooks is a web browser–based source editing tool (Web IDE) that provides convenience for writing code.

Supported notebook images

The following notebook images are supported. If the required ML framework is not available, you can use a custom Docker image.

info

The image registry endpoint is mlops.kr-central-2.kcr.dev/kc-kubeflow-registry/(image-name).

• For example, to pull the image kc-jupyter-pytorch-full:natl.py311.cu130.1a, use
  mlops.kr-central-2.kcr.dev/kc-kubeflow-registry/kc-jupyter-pytorch-full:natl.py311.cu130.1a.

Supported notebook information

Image name	Python version	Framework (version)	GPU support
kc-kubeflow-registry/kc-jupyter-scipy:natl.py311.cpu.1a	python 3.11.11	-	X
kc-kubeflow-registry/kc-jupyter-scipy:natl.py311.cu130.1a	python 3.11.11	-	O
kc-kubeflow-registry/kc-jupyter-tensorflow:natl.py311.cpu.1a	python 3.11.11	tensorflow(2.21.0)	X
kc-kubeflow-registry/kc-jupyter-tensorflow:natl.py311.cu130.1a	python 3.11.11	tensorflow(2.21.0)	O
kc-kubeflow-registry/kc-jupyter-tensorflow:natl.py312.cu128.1b	python 3.12.3	tensorflow(2.17.0)	O
kc-kubeflow-registry/kc-jupyter-pytorch:natl.py311.cpu.1a	python 3.11.11	pytorch(2.5.1)	X
kc-kubeflow-registry/kc-jupyter-pytorch:natl.py311.cu130.1a	python 3.11.11	pytorch(2.10.0.dev20251202)	O
kc-kubeflow-registry/kc-jupyter-pytorch:natl.py311.cu130.1b	python 3.11.11	pytorch(2.9.1)	O
kc-kubeflow-registry/kc-jupyter-tensorflow-full:natl.py311.cpu.1a	python 3.11.11	tensorflow(2.21.0)	X
kc-kubeflow-registry/kc-jupyter-tensorflow-full:natl.py311.cu130.1a	python 3.11.11	tensorflow(2.21.0)	O
kc-kubeflow-registry/kc-jupyter-tensorflow-full:natl.py312.cu128.1b	python 3.12.3	tensorflow(2.17.0)	O
kc-kubeflow-registry/kc-jupyter-pytorch-full:natl.py311.cpu.1a	python 3.11.11	pytorch(2.5.1)	X
kc-kubeflow-registry/kc-jupyter-pytorch-full:natl.py311.cu130.1a	python 3.11.11	pytorch(2.10.0.dev20251202)	O
kc-kubeflow-registry/kc-jupyter-pytorch-full:natl.py311.cu130.1b	python 3.11.11	pytorch(2.9.1)	O
kc-kubeflow-registry/kc-jupyter-pyspark-tensorflow:natl.py311.cpu.1a	python 3.11.11	tensorflow(2.21.0)	X
kc-kubeflow-registry/kc-jupyter-pyspark-tensorflow:natl.py311.cu130.1a	python 3.11.11	tensorflow(2.21.0)	O
kc-kubeflow-registry/kc-jupyter-pyspark-tensorflow:natl.py312.cu128.1b	python 3.12.3	tensorflow(2.17.0)	O
kc-kubeflow-registry/kc-jupyter-pyspark-pytorch:natl.py311.cpu.1a	python 3.11.11	pytorch(2.5.1)	X
kc-kubeflow-registry/kc-jupyter-pyspark-pytorch:natl.py311.cu130.1a	python 3.11.11	pytorch(2.10.0.dev20251202)	O
kc-kubeflow-registry/kc-jupyter-pyspark-pytorch:natl.py311.cu130.1b	python 3.11.11	pytorch(2.9.1)	O
kc-kubeflow-registry/kc-codeserver-pytorch:natl.py311.cpu.1a	python 3.11.11	pytorch(2.5.1)	X
kc-kubeflow-registry/kc-codeserver-pytorch:natl.py311.cu130.1a	python 3.11.11	pytorch(2.10.0.dev20251202)	O
kc-kubeflow-registry/kc-codeserver-pytorch:natl.py311.cu130.1b	python 3.11.11	pytorch(2.9.1)	O
kc-kubeflow-registry/kc-codeserver-tensorflow:natl.py311.cpu.1a	python 3.11.11	tensorflow(2.21.0)	X
kc-kubeflow-registry/kc-codeserver-tensorflow:natl.py311.cu130.1a	python 3.11.11	tensorflow(2.21.0)	O
kc-kubeflow-registry/kc-codeserver-tensorflow:natl.py312.cu128.1b	python 3.12.3	tensorflow(2.17.0)	O
kc-kubeflow-registry/kc-jupyter-llm-dev-kit:natl.py311.cu130.1a	python 3.11.11	pytorch(2.10.0.dev20251202)	O
kc-kubeflow-registry/kc-jupyter-llm-dev-kit:natl.py311.cu130.1b	python 3.11.11	pytorch(2.9.1)	O

info

If you select Custom Image, you can specify an image using a Docker Hub or public registry URL.

• If the image is not based on Jupyter Lab, the notebook may not be created properly.

Dockerfile example for building a notebook

ARG BASE_IMG=<mlops.kr-central-2.kcr.dev/kc-kubeflow-registry/kc-jupyter-pytorch-full:natl.py311.cu130.1a>
FROM $BASE_IMG

ARG PYTHON_VER=38
ARG FRAMEWORK_TYPE=38

ENV DEBIAN_FRONTEND=noninteractive

USER $NB_UID

#
# *** YOUR CUSTOM DOCKER CODE HERE ***
#

COPY --chown=jovyan:users requirements.txt /tmp/requirements.txt

RUN python3 -m pip install --upgrade pip --quiet \
    && python3 -m pip install -r /tmp/requirements.txt --no-cache-dir \
    && rm -f /tmp/requirements.txt
# INSTALLABLE PYTHON PACKAGES IN requirements.txt

Example of creating requirements.txt

    ## *** YOUR INSTALLABLE PACKAGES ***
    ## ex) tensorflow==2.5.2   

info

BASE_IMG: Refer to Supported notebook images.

Create notebook instance

Create a notebook instance with the desired specifications to build a computing environment required for running ML code and processing data.

1. Access the login page of the National AI computing resource support portal and move to the Kubeflow page.

2. Select the Notebooks tab located on the left. Access the Notebooks tab from the dashboard

3. Click the [New Notebook] button located at the top right.

4. On the New Notebook screen, enter the required information and click the [LAUNCH] button to create the notebook instance.

   Create notebook

   Item	Category	Description
   Name	Name	Used to identify the notebook instance on the dashboard
   	Namespace	Kubernetes namespace where the notebook instance is created
   Docker Image	Custom Image	If Custom Image is selected, you can specify an image using Docker Hub or a public registry URL
   	Image	Select one of JupyterLab / VisualStudio Code / RStudio, then select the image to install
   CPU / RAM	Requested CPUs	2 - Number of CPU cores, specifies the amount of CPU resources used by the notebook instance
   	Requested memory in Gi	4 - Memory resource unit (GiB), specifies the amount of memory resources used by the notebook instance
   GPU	Number of GPUs	Optional - Number of GPUs, specifies the amount of GPU resources used by the notebook instance
   Workspace Volume	New Volume(default)	Create and add a new volume to be mounted to the directory (Storage class types)
   	Add existing volume	Select and add an existing volume to be mounted to the directory
   Data Volumes	Add new volume	Create and add a new volume to be mounted inside the notebook instance
   	Add existing volume	Select and add an existing volume to be mounted inside the notebook instance
   Enable SSH Connection	Enable if SSH access to the notebook is required SSH connection guide

Access notebook instance

1. Access the login page of the National AI computing resource support portal and move to the Kubeflow page.

2. Select the Notebooks tab located on the left.

3. In the list of created Notebooks, click the [CONNECT] button for the notebook you want to access.

4. When you click the [CONNECT] button, a new tab opens and the Jupyter Notebook interface for the notebook instance is displayed.

   • You can open notebooks, write code, or run machine learning models.

   Jupyter Notebook interface

5. You can use the notebook instance through the Terminal.

   Click terminal

Stop notebook instance

Stop a notebook instance mainly for resource savings, security, cost reduction, and environment management. If a notebook instance continues running, resources cannot be allocated to other tasks, and security issues and increased costs may occur.

info

No charges are incurred for stopped notebook instances.

1. Access the login page of the National AI computing resource support portal and move to the Kubeflow page.

2. Select the Notebooks tab located on the left.

   

3. In the Notebooks list, click the ◼ (Stop) button in the row of the notebook you want to stop.

   • When the stop confirmation modal appears, click the [STOP] button.

   Confirm notebook stop

4. In the Notebooks list, check that the status (Status) of the stopped notebook is correctly set to stopped.

   Verify notebook stop

Delete notebook instance

Delete a notebook instance mainly for resource reclamation, security, cost reduction, and environment management. When a notebook instance is deleted, the associated resources are reclaimed and can be allocated to other tasks. For more details, refer to the Kubeflow > Notebook official documentation.

caution

Deleted notebook instance names cannot be reused in the future.

1. Access the login page of the National AI computing resource support portal and move to the Kubeflow page.

2. Select the Notebooks tab located on the left.

   

3. In the Notebooks list, click the delete (trash) button in the row of the notebook you want to delete.

   • When the delete confirmation modal appears, click the [DELETE] button.

   Notebook delete notice

4. In the Notebooks list, verify that the deleted notebook no longer exists.

   Verify notebook deletion

Upload files to notebook

1. Upload files through the browser.

   Upload files via the Jupyter Notebook console file browser

2. After the upload is complete, you can view the contents of the uploaded file in the right pane.

   Example file upload completed

Download files from notebook

Click the [Download] button from the [File] tab at the top to download files.

Tensorboards

Tensorboard is a tool for visualizing the results of Notebooks or Pipelines. For basic usage, select the storage where the logs to be used by Tensorboard are stored, then enter the mounted path value to read and visualize the data.

Create tensorboard

1. Click the [New TensorBoard] button in the TensorBoards tab.

   

2. Enter the Tensorboard name and select the storage (PVC) where the logs are stored.

   

3. After selecting the PVC menu, choose the bucket that contains the files to review from the General bucket tab, enter the mount path value, and click the [Create] button.

Volumes

Create volume

Volume is the space where actual data is stored. Both input and output data can be stored in Volumes.

info

Created Volumes can be used as persistent volumes required when creating Notebooks and Tensorboards.

1. Move to the Volumes tab on the dashboard and click the [New Volume] button at the top. Volume dashboard screen

2. On the New Volume screen, enter the required information, select the Storage Class to use, and click the [Create] button to create the volume.

Storage class types

Select the Storage Class according to data importance and retention requirements.

• gpfs (For AI training and data): A high-performance parallel file system that provides the fastest I/O speed.
• local-pv (For temporary data storage): Uses the local disk of a physical server, providing fast speed, but data may be lost or become inaccessible when the notebook restarts or the server changes.

Experiments

Experiment is a type of specification (definition) that defines which model to run with which input data, and provides AutoML and KFP methods.

Experiments (AutoML)

AutoML is a tool that automates repeated experiments to improve prediction accuracy and performance of machine learning models, and provides AutoML functionality using Katib.

Experiments (AutoML) dashboard screen

Pipelines

Create pipeline

Pipelines is a feature that allows you to write a series of processes—such as specifying datasets, training models, and evaluating models—in code and execute them sequentially or in parallel at once.
In other words, it provides a visual console to handle and manage machine learning workflows such as preprocessing, training, and serving.

For detailed instructions on pipeline creation, refer to the Kubeflow > Kubeflow Pipeline > Quick Start documentation.

1. From the dashboard, move to the Pipelines tab and click the [Upload pipeline] button.

   Access the Pipelines tab from the dashboard

2. Upload the pipeline manifest file or import a Package URL, then click the [Create] button to create the pipeline.

   Create pipeline

Delete pipeline

1. From the dashboard, select the Pipelines tab.
2. In the list view, select the pipeline to delete, then click the [Delete] button at the top right to delete it.