Fix tests

genienlp/predict.py

@@ -135,7 +135,7 @@ def parse_argv(parser):
     parser.add_argument(
         '--val_batch_size',
         nargs='+',
-        default=[4000],
+        default=None,
         type=int,
         help='Batch size for validation corresponding to tasks in val tasks',
     )
@@ -171,10 +171,9 @@ def parse_argv(parser):
         default=[0],
         help='ngrams of this size cannot be repeated in the output. 0 disables it.',
     )
-    parser.add_argument('--max_output_length', default=150, type=int, help='maximum output length for generation')
+    parser.add_argument('--max_output_length', type=int, help='maximum output length for generation')
     parser.add_argument(
         '--min_output_length',
-        default=3,
         type=int,
         help='maximum output length for generation; '
         'default is 3 for most multilingual models: BOS, language code, and one token. otherwise it is 2',
@@ -325,9 +324,9 @@ def check_args(args):
     if not args.pred_tgt_languages:
         setattr(args, 'pred_tgt_languages', [args.eval_tgt_languages])
     
-    if args.is_hf_model and (not args.pred_src_languages or not args.model):
+    if args.is_hf_model and (not args.pred_src_languages or not args.model or not args.min_output_length or not args.max_output_length or not args.val_batch_size):
         # because in for HF models we are not getting these values from genienlp's training script
-        raise ValueError('You need to specify --pred_languages and --model when directly loading a HuggingFace model.')
+        raise ValueError('You need to specify --pred_languages, --model, --min_output_length, --max_output_length and --val_batch_size when directly loading a HuggingFace model.')
 
     if len(args.task_names) != len(args.pred_src_languages):
         raise ValueError('You have to define prediction languages for each task in the same order you provided the tasks.')

genienlp/util.py

@@ -585,7 +585,11 @@ def have_multilingual(task_names):
     return any(['multilingual' in name for name in task_names])
 
 
-def load_config_file_to_args(args) -> bool:
+def load_config_file_to_args(args):
+    if not hasattr(args, 'is_hf_model'):
+        # --is_hf_model might not exist if this function is called by anything other than predict.py
+        setattr(args, 'is_hf_model', False)
+        
     if args.is_hf_model:
         # no config file found, treat `args.path` as a model name on HuggingFace model hub
         args.pretrained_model = args.path

tests/test_e2e_dialogues.sh

@@ -26,7 +26,8 @@ do
     --train_iterations 4 \
     --min_output_length 2 \
     --save $workdir/model_$i \
-    --data $SRCDIR/dataset/bitod
+    --data $SRCDIR/dataset/bitod \
+    ${hparams[i]}
 
   # greedy prediction
   genienlp predict \

tests/test_paraphrasing.sh

@@ -42,6 +42,9 @@ for model in \
     --embeddings $EMBEDDING_DIR \
     --pred_languages en \
     --model TransformerSeq2Seq \
+    --min_output_length 1 \
+    --max_output_length 150 \
+    --val_batch_size 100 \
     --is_hf_model
 
   # check if result file exists

tests/test_sequence_classification.sh

@@ -20,7 +20,16 @@ genienlp train \
   --num_print 0
 
 # greedy prediction
-genienlp predict --tasks ood_task --evaluate valid --pred_set_name eval --path $workdir/model --overwrite --eval_dir $workdir/model/eval_results/ --data $SRCDIR/dataset/ood/ --embeddings $EMBEDDING_DIR  --val_batch_size 200
+genienlp predict \
+  --tasks ood_task \
+  --evaluate valid \
+  --pred_set_name eval \
+  --path $workdir/model \
+  --overwrite \
+  --eval_dir $workdir/model/eval_results/ \
+  --data $SRCDIR/dataset/ood/ \
+  --embeddings $EMBEDDING_DIR \
+  --val_batch_size 200
 
 # check if result file exists
 if test ! -f $workdir/model/eval_results/valid/ood_task.tsv ; then

tests/test_token_classification.sh

@@ -20,7 +20,8 @@ do
     --val_batch_size 200 \
     --train_iterations 4 \
     --save $workdir/model_$i \
-    --data $SRCDIR/dataset/cross_ner/
+    --data $SRCDIR/dataset/cross_ner/ \
+    $hparams
 
   # greedy prediction
   genienlp predict \
@@ -66,7 +67,8 @@ do
     --val_batch_size 100 \
     --train_iterations 4 \
     --save $workdir/model_$i \
-    --data $SRCDIR/dataset/cross_ner/
+    --data $SRCDIR/dataset/cross_ner/ \
+    $hparams
 
   # greedy prediction
   genienlp predict \