HPC Job Template

Creating HPC job based on template can benefits in rapid Job deployment, user can upload a Linux bash script file (.sh) or create a brand new .sh file on scratch or its sub-directory with specified format and used the file as template (A simple Slurm Job template created for each new user by default, path: /home/<EdUHK User ID>/job_template)

Example for HPC Job using Environment Module (Using CPU only)

#!/bin/bash
#SBATCH --job-name=hello_world ## Job Name
#SBATCH --partition= shared_cpu ## Partition for Running Job
#SBATCH –-constraint=CPU_FRQ:2.9GHz|CPU_GEN:XEON-5 ## Select only nodes with CPUs provides 2.9GHz Clock or 5th generation XEON CPUs
#SBATCH --nodes=1 ## Number of Compute Node
#SBATCH --ntasks-per-node=1 # Number of Task per Compute Node
#SBATCH --cpus-per-task=15 ## Number of CPU per task
#SBATCH --time=60:00 ## Job Time Limit (i.e. 60 Minutes)
#SBATCH --mem=60GB ## Total Memory for Job
#SBATCH --output=./%x%j.out ## Output File Path
#SBATCH --error=./%x%j.err ## Error Log Path
#SBATCH --mail-type=ALL ## Enable Email Notification For Job Status (Optional)


## Initiate Environment Module
source /usr/share/modules/init/profile.sh

## Reset the Environment Module components
module purge

## Load Module
module load python/3.10

## Export User Local Package Path (If Extra Python Packages Installed)
export PATH="/home/$USER/.local/bin:$PATH"

## Run User Command
python3 /scratch/$USER/hello_world.py

## Clear Environment Module components
module purge

Example for HPC Job using user’s environment created by Anaconda (Using CPU only)

#!/bin/bash
#SBATCH --job-name=hello_world ## Job Name
#SBATCH --partition=shared_cpu ## Partition for Running Job
#SBATCH –-constraint=NO_GPU&CPU_SKU:6542Y ## Select only nodes with Intel Xeon Gold 6542Y CPU chips and no GPU installed
#SBATCH --nodes=1 ## Number of Compute Node
#SBATCH --ntasks-per-node=1 # Number of Task per Compute Node
#SBATCH --cpus-per-task=15 ## Number of CPU per task
#SBATCH --time=60:00 ## Job Time Limit (i.e. 60 Minutes)
#SBATCH --mem=60GB ## Total Memory for Job
#SBATCH --output=./%x%j.out ## Output File Path
#SBATCH --error=./%x%j.err ## Error Log Path
#SBATCH --mail-type=ALL ## Enable Email Notification For Job Status (Optional)


## Initiate Environment Module
source /usr/share/modules/init/profile.sh

## Reset the Environment Module components
module purge<br>

## Load Anaconda Module
module load anaconda

## Run user command (“conda activate” command is not available in script, use absolute path for the environment instead, an ${CONDA_ENV_PATH} variable is available for representing “/home/$USER/.conda/envs” path)

${CONDA_ENV_PATH}/&lt;environment name&gt;/bin/python3 /scratch/$USER/hello_world.py

Example for HPC Job using Singularity container (using GPU resources)

#!/bin/bash<br>
#SBATCH --job-name=hello_world ## Job Name
#SBATCH --partition=shared_gpu_h20 ## Partition for Running Job
#SBATCH –-constraint=CPU_MNF:INTEL&GPU_GEN:HOPPER ## Select only nodes with Intel CPUs and HOPPER architecture GPU (used with --gres)
#SBATCH --nodes=1 ## Number of Compute Node
#SBATCH --ntasks-per-node=1 # Number of Task per Compute Node
#SBATCH --cpus-per-task=10 ## Number of CPU per task
#SBATCH --time=60:00 ## Job Time Limit (i.e. 60 Minutes)
#SBATCH --mem=40GB ## Total Memory for Job
#SBATCH --gres=gpu:h20:2 ## Number of GPUs (i.e. Request 2 x H20 GPUs)
#SBATCH --output=./%x%j.out ## Output File Path
#SBATCH --error=./%x%j.err ## Error Log Path
#SBATCH --mail-type=ALL ## Enable Email Notification For Job Status (Optional)

## Initiate Environment Module
source /usr/share/modules/init/profile.sh

## Reset the Environment Module components
module purge

## Load Singularity Module
module load singularity

## Run user command in container (Python container – python.sif as example)
singularity run --bind /scratch/$USER /scratch/$USER/python.sif python3 /scratch/$USER/hello_world.py

Create Template (Web Interface Feature)

1. Upload or create Linux bash script file (.sh)
2. Go to “Jobs”  “Job Composer”
3. Go to “Templates”
4. Click “New Template”
5. Fill-in the form, “Copy Path” function in “Files” can be used while user located the template file directory
6. Save the template, and which is available on the platform