Bash script for launching distributed TensorFlow training in several processes.

run.sh

# Simple script for experiments with dsitributed TensorFlow.
# Requires jq utility for JSON parsing. Example JSON file can be found on the gist.
# Args:
#     name of script to run, example: train.py
#     file with cluster configuration, example: hosts.json
#!/bin/bash

script=$1
hosts=$2

for job in `jq -cr 'keys[]' $hosts`
do
  task=0
  for hostinfo in `jq -c ".$job[]" $hosts`
  do
    host=`echo $hostinfo | $jq -r '.host'`
    # run script locally
    python $script --job $job --task $task --hosts $hosts &
    # or some command to run script remotely
    # ssh -f user@$host "..."
    ((task++))
  done
done