Example of launcher script with Horovod

[guillaumeguy]

Hi there:

Is there a good script template to launch a training job with Horovod? I couldn't find one in ./scripts/ and the Horovod documentation use horovod.run (is it what you would recommend??).

Thanks!

https://github.com/horovod/horovod/blob/master/examples/pytorch/pytorch_mnist.py#L253

[guillaumeguy]

This worked for me:

if [ -z "$PYTHONPATH" ]; then
    export PYTHONPATH=$PWD
else
    PYTHONPATH=$PYTHONPATH:$PWD
    export PYTHONPATH
fi

echo "PYTHONPATH" $PYTHONPATH

python training/main.py \
--save-frequency 10 \
--save-most-recent \
--train-data "$TRAINING_DATA"  \
--val-data "$VAL_DATA"  \
...