Guidelines to run Keras DL model on Biowulf GPU nodes using singularity.
Export the environment variable SINGULARITY_BINDPATH in your ~/.bashrc file.
export SINGULARITY_BINDPATH="/gpfs,/gs2,/gs3,/gs4,/gs5,/gs6,/spin1,/data,/scratch,/fdb,/lscratch"
This will automatically bind all the directories on Biowulf to the singularity image we use on Biowulf.
Clone the CANDLE benchmarks, and supervisor in your data directory:
candle_dir=/data/`whoami`/candle mkdir $candle_dir cd $candle_dir git clone https://github.com/ECP-CANDLE/Benchmarks.git git clone https://github.com/ECP-CANDLE/Supervisor.git
To run any of the benchmarks on GPU nodes interactively, get a GPU node interactively using:
sinteractive --gres=gpu:k80:1 --mem=25G
once on the node run:
module load singularity singularity exec --nv /data/classes/candle/candle-gpu.img <benchmark-python-script>
For example, to run P1B1, run:
singularity exec --nv /data/classes/candle/candle-gpu.img python /PATH/TO/Benchmarks/Pilot1/P1B1/p1b1_baseline_keras2.py
singularity shell --nv /data/classes/candle/candle-gpu.img