yetulaxman/ai4spec

Deploying AI4Spec Notebooks at CSC

Course participants are expected to have a user account at CSC and be a member of a project that has access to the Puhti service. The project can be a course project created by organisers. The familiarity with the following main topics help organisers get started smoothly with notebooks installation:

• A brief primer on using Puhti computing environment
• Preparing custom Jupyter notebook for course
• Creating a course environment/module(s) on Puhti Web Interface
• Accessing notebook via Puhti web interface
• Useful CSC documentation

A Brief Primer on Using Puhti Computing Environment

• Module system: CSC uses module system to manage complex application stack in supercomputing environment. Applications installed as modules can easily be used in both interactive and batch jobs. The detailed instructions on using modularised applications can be found in CSC documentation pages as well as a CSC course page
• Disk areas: CSC supercomputers have three main disk areas namely home, projappl and scratch which are visible to all compute and login nodes. Each disk area has its own dedicated use and comes with quota limits on the size of disk pace and the number of files. Default quotas and their specific use can be found in CSC user documentation
• Custom installations: You can install own software on CSC supercomputers if you cannot find it from the list of pre-installed applications or using module spider command on Puhti terminal. Typically, one can download the source code of the software, compiles the code, and installs to a location where the user has write-access, e.g. the project's /projappl directory. More about installations can be found on CSC documentation page and a CSC course pages. Puhti also supports containerised installations
• Puhti web interface: You can use web interface for Puhti to access Puhti supercomputer. The web interface greatly eases the use of complex applications that have graphical user interfaces among other uses. Read more information about the web interface on CSC documentation page

Preparing a Custom Jupyter Notebook for Course

CSC provisions popular python environements as ready-to-use notebooks. You can however customise a python environement to meet your own needs. A custom Jupyter notebook for AI4Spec notebooks for course can be prepared in command-line environment of Puhti and accessed it through Puhti web interface. The customisation of notebook typically involves the following procedure:

Installing Necessary Python Packages to Projappl Directory Using tykky

Conda installations should not be done directly on Puhti. Tykky wrapper tool instead be used to install python packages in setting up your compute environment. The wrapper tool installs applications inside of a singularity container and thus facilitates better performance in terms of faster startup times, reduced IO load, and reduced number of files on parallel filesystems.

Here is an example of tykky-based custom installation of AI4Spec noteboks for course (make sure to edit with correct CSC project name and user name as needed):

# start an interactive session once you are in login node
sinteractive -c 4 -m 30000 -d 100  # this command requests a compute node with 4 cores, 30 GB memory and 100 GB local disk space; change settings as needed
# load needed packages 
module load git   # git command is not available by default on interactive nodes
module load purge  # clean environment 
module load tykky # load tykky wrapper 
# install python libraries using tykky wrapper tool; make sure to use proper project/username
mkdir -p /projappl/<project>/$USER && mkdir -p /projappl/<project>/$USER/AI4Spec
conda-containerize new --prefix  /projappl/<project>/$USER/AI4Spec AI4Spec_python3.7.yml     # install basic packages using .yml
export PATH="/projappl/<project>/$USER/AI4Spec/bin:$PATH"
conda-containerize update /projappl/<project>/$USER/AI4Spec/  --post-install requirements_AI4Spec.txt   # update package list 

In the above example, Tykky installs a basic setup (as listed in the file, AI4Spec_python3.7.yml) first and then updates all python packages (as listed in the file, requirements_AI4Spec.txt) to the directory '/projappl/project_xxxx/AI4Spec'. Please note that you have to add the bin directory of installation to the $PATH variable before start using the installed environment (i.e., export PATH="/projappl/project_xxxx/$USER/AI4Spec/bin:$PATH").

Creating a Course Environment Module on Puhti Web Interface

One has to create a course environments (modules) in the directory /projappl/project_xxxx/www_puhti_modules/ to be able to see a course module in Puhti web insterface ("under Jupyter for courses") . The www_puhti_modules directory can be created if it does not exist. Please note that module_name has to be unique accross the project members.

The two files needed for setting up the course modules are:

• a <<module_name>>.lua file that defines the module that sets up the Python environment. Only files containing the text Jupyter will be visible in the app.
• a <<module_name>>-resources.yml that defines the default resources used for Jupyter.

For AI4Spec notebooks, the above mentioned two files for each tutorial (i.e., .lua and .yaml files) are made available in config folder in the GitHub repository we cloned above. So just copy them over to appropriate place under /projappl folder as below:

mkdir -p /projappl/project_xxxx/www_puhti_modules && cp AI4Spec_tutorial1-resources.yml AI4Spec_tutorial1.lua /projappl/project_xxxx/www_puhti_modules

# Edit the correct project number and actual username (in AI4Spec_tutorial1.lua file) in the copied files in /projappl/project_xxxx/www_puhti_modules.

Once you have set up modules as instructed above, a module named, AI4Spec_tuotiral1 is visible in "Jupyter for courses" app in the Puhti web interface

Accessing Notebook via Puhti Web Interface

1. Login to Puhti web interface
2. Login with your CSC account (or Haka/Virtu) (Users should have accepted Puhti service in myCSC page under a course ( or own) project before using this service). Login page is as shown below:

3. Once login is successfull, you can click "Interactive Sessions" on the top menu bar and then click "Jupyter for courses" (or directly from pinned apps on landing page). Once the app is launched, you can see the different fields to fill in before launching a job. For this course, select the "course Project" and Working directory corresponding to course project. Then you will be able to see "AI4Spec_tutorial" module under the "Course module" field. You can then launch Jupyter notebook whick will be launched in the interaction partition by default.

‼️ Please note that if you reserve any GPU resources for course, the reservations field is visible ONLY during the reservation time.

4. Upon successful launching a job, you can click on "connect to Jupyter" to see the course notebook corresponidng to AI4Spec_turoial.

Useful CSC Documentation

• Jupyter for course
• Tykky containerisation