cnn-gpu-job.lsf

#BSUB -P trn018
##BSUB -q debug
#BSUB -W 0:30
#BSUB -nnodes 1
#BSUB -J cnn-gpu-pretrained-1
#BSUB -o logs/cnn-gpu-pretrained-1-%J.o
#BSUB -e logs/cnn-gpu-pretrained-1-%J.e

module load open-ce
export TORCH_HOME=/gpfs/wolf/world-shared/trn018/ai-training-series/ai_at_scale/TH
export http_proxy=http://proxy.ccs.ornl.gov:3128/
export https_proxy=https://proxy.ccs.ornl.gov:3128/

export OMP_NUM_THREADS=4
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5

time jsrun -n 1 --tasks_per_rs 1 --cpu_per_rs 42 --gpu_per_rs 1 --rs_per_host 1 --latency_priority CPU-CPU --launch_distribution packed --bind packed:1 python3 -u cnn_classifier_gpu.py