fix bug in pt/translation example (#1128)

Author: xin3he

examples/pytorch/nlp/huggingface_models/translation/quantization/ptq_dynamic/eager/run_benchmark.sh

@@ -52,8 +52,7 @@ function init_params {
 
 # run_benchmark
 function run_benchmark {
-    extra_cmd=''
-
+    extra_cmd='None'
     if [[ ${mode} == "accuracy" ]]; then
         mode_cmd=" --accuracy_only"
     elif [[ ${mode} == "benchmark" ]]; then
@@ -65,10 +64,9 @@ function run_benchmark {
 
     if [ "${topology}" = "t5_WMT_en_ro" ];then
         model_name_or_path='t5-small'
-        extra_cmd="--source_lang en --target_lang ro --dataset_name wmt16 --dataset_config_name ro-en"
+        extra_cmd='translate English to Romanian: '
     elif [ "${topology}" = "marianmt_WMT_en_ro" ]; then
         model_name_or_path='Helsinki-NLP/opus-mt-en-ro'
-        extra_cmd="--source_lang en --target_lang ro --dataset_name wmt16 --dataset_config_name ro-en"
     fi
 
     if [[ ${int8} == "true" ]]; then
@@ -82,9 +80,12 @@ function run_benchmark {
         --predict_with_generate \
         --per_device_eval_batch_size ${batch_size} \
         --output_dir ${tuned_checkpoint} \
-        --source_prefix "translate English to Romanian: " \
+        --source_lang en \
+        --target_lang ro \
+        --dataset_name wmt16 \
+        --dataset_config_name ro-en\
         ${mode_cmd} \
-        ${extra_cmd}
+        --source_prefix "$extra_cmd"
 }
 
 main "$@"

examples/pytorch/nlp/huggingface_models/translation/quantization/ptq_dynamic/eager/run_tuning.sh

@@ -37,18 +37,17 @@ function init_params {
 
 # run_tuning
 function run_tuning {
-    extra_cmd=''
+    extra_cmd='None'
     batch_size=16
     model_type='bert'
 
     if [ "${topology}" = "t5_WMT_en_ro" ];then
         model_name_or_path='t5-small'
         model_type='t5'
-        extra_cmd="--source_lang en --target_lang ro --dataset_name wmt16 --dataset_config_name ro-en"
+        extra_cmd='translate English to Romanian: '
     elif [ "${topology}" = "marianmt_WMT_en_ro" ]; then
         model_name_or_path='Helsinki-NLP/opus-mt-en-ro'
         model_type='marianmt'
-        extra_cmd="--source_lang en --target_lang ro --dataset_name wmt16 --dataset_config_name ro-en"
     fi
 
     sed -i "/: bert/s|name:.*|name: $model_type|g" conf.yaml
@@ -61,10 +60,13 @@ function run_tuning {
         --predict_with_generate \
         --per_device_eval_batch_size ${batch_size} \
         --output_dir ${tuned_checkpoint} \
-        --source_prefix "translate English to Romanian: " \
+        --source_lang en \
+        --target_lang ro \
+        --dataset_name wmt16 \
+        --dataset_config_name ro-en\
         --tune \
         --overwrite_output_dir \
-        $extra_cmd
+        --source_prefix "$extra_cmd"
 }
 
 main "$@"