Created
March 3, 2021 21:12
-
-
Save Spotlight0xff/148d5d5a5fe9210a1ff9d70eec63bac3 to your computer and use it in GitHub Desktop.
Slurm file for training a Transducer model on 8 GPUs
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#!/usr/local_rwth/bin/zsh | |
#SBATCH --mail-user=andre.merboldt@rwth-aachen.de | |
#SBATCH --mail-type=ALL | |
#SBATCH --nodes=1 | |
#SBATCH --ntasks-per-node=8 | |
#SBATCH --cpus-per-task=2 | |
#SBATCH --time=120:00:00 # 5 days | |
#SBATCH --output=log/train-multi-gpu-job.%J.log # stdout/stderr file | |
#SBATCH --partition=dgx2 # DGX2 (has 16 GPUs) | |
#SBATCH --account=supp0003 # substitute appropriate group here | |
#SBATCH --job-name=train-rnnt | |
#SBATCH --gres=gpu:8 | |
source base/venv/bin/activate | |
source settings.sh | |
module unload intelmpi | |
module load openmpi/4.0.3 | |
module load cuda/101 | |
module load cudnn/7.6.5 | |
module load gcc/7 | |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/work/am540506/downloads/nccl_2.8.3-1+cuda10.1_x86_64/lib | |
export HOROVOD_TIMELINE=log/horovod.timeline | |
export I_MPI_DEBUG=3 | |
#print some debug informations... | |
echo; export; echo; nvidia-smi; echo | |
mpirun -np 8 \ | |
-bind-to none -map-by slot \ | |
-x TMPDIR -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x HOROVOD_TIMELINE -x DEBUG \ | |
-mca pml ob1 -mca btl ^openib \ | |
python3 returnn/rnn.py config-train/$model.config ++use_horovod 1 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment