Added a MirroredStrategy to train models on multiple GPUs if they are available

run_deepof_experiments.sh

@@ -8,5 +8,6 @@ snakemake --snakefile deepof_experiments.smk --forceall --rulegraph | dot -Tpdf
 module load anaconda/3
 module load tensorflow/gpu/2.4.0
 module load cuda/10.2
+module load cudnn/8.0.4
 
 pipenv run snakemake --snakefile deepof_experiments.smk deepof_experiments -j 20 --latency-wait 15 --cluster-config cluster.json --cluster "sbatch --time={cluster.time} --mem={cluster.mem} -o {cluster.stdout} -e {cluster.stderr} --job-name={cluster.jobname} --cpus-per-task={cluster.cpus} --constraint="gpu" --gres=gpu:gtx980:2" > deepof_experiments.out 2> deepof_experiments.err