Testing a fairseq Translation Model

test.sh

#!/usr/bin/env bash
#
# Adapted from https://github.com/facebookresearch/MIXER/blob/master/prepareData.sh
usage() {
    echo "usage: <command> options:<e|p|b|a>"
    echo "-e exp1 -p eng-brx -b 5000 -a transformer"
}
no_args="true"
while getopts e:l:b:a: flag; do
    case "${flag}" in
    e) experiment=${OPTARG} ;;
    l) lang_pair=${OPTARG} ;;
    b) bpe_token=${OPTARG} ;;
    a) architecture=${OPTARG} ;;
    *)
        usage
        exit
        ;;
    esac
    no_args="false"
done
[[ "$no_args" == "true" ]] && {
    usage
    exit 1
}
export exp=$experiment
export bpe_count=$bpe_token
langs=(${lang_pair//-/ })
export src=${langs[0]}
export tgt=${langs[1]}
export arch=${architecture:-"transformer"}
python test.py ${exp}/checkpoints/${arch}.${src}-${tgt}.${bpe_count} ${exp}/data-bin/${arch}.${src}-${tgt}.${bpe_count} data/bpe_data/bpe_data_${bpe_count}/bpe_data_${bpe_count}