decode_chunk.sh

#! /bin/bash

chunk_input_dir=$1
chunk_output_dir=$2
chunk_prefix=$3
chunk_index=$4

model_paths=$5
batch_size=$6

OMP_NUM_THREADS=3 python -m sockeye.translate \
            -i $chunk_input_dir/$chunk_prefix"$chunk_index" \
            -o $chunk_output_dir/$chunk_prefix"$chunk_index" \
            -m $model_paths \
            --beam-size 10 \
            --length-penalty-alpha 1.0 \
            --device-ids 0 \
            --batch-size $batch_size

decode_parallel.sh

#! /bin/bash

base=[path on your machine]

# vars you need to define:

# corpus, e.g. "test"
corpus=test
# model_name, e.g. "baseline
model_name=baseline
# model_paths
$base/models/baseline
# batch_size
64
# chunk_size (in sentences)
100000

data=$base/data
scripts=$base/scripts

translations=$base/translations

src=en
trg=de

mkdir -p $translations
mkdir -p $translations/$model_name

chunk_prefix="$corpus.bpe.$model_name.chunk."
chunk_input_dir=$translations/$model_name/chunk_inputs
chunk_output_dir=$translations/$model_name/chunk_outputs
chunk_log_dir=$translations/$model_name/chunk_logs

mkdir -p $chunk_input_dir
mkdir -p $chunk_output_dir
mkdir -p $chunk_log_dir

# splitting input file into chunks

zless $data/$corpus.$src | split -d -l $chunk_size -a 3 - $chunk_input_dir/$chunk_prefix

# get number of chunk files generated

num_chunks=`ls $chunk_input_dir | wc -l`

echo "Number of chunks found: $num_chunks"

# translating individual chunks

for chunk_index in $(seq -f "%03g" 0 $(($num_chunks - 1))); do
    . $scripts/decode_chunk.sh \
        $chunk_input_dir $chunk_output_dir $chunk_prefix $chunk_index $model_paths $batch_size
done

# move logs out of the way

mv $chunk_output_dir/*.log $chunk_log_dir/

# concatenating results

cat $chunk_output_dir/$chunk_prefix* > $translations/$model_name/$corpus.bpe.$model_name.$trg