Fix dataset validation and label mapping

app.py

@@ -270,8 +270,8 @@ with gr.Blocks(theme=theme) as iface:
                 placeholder="tweet_eval",
             )
         with gr.Row():
-            dataset_config_input = gr.Dropdown(['default'], value=['default'], label='Dataset Config', visible=False)
-            dataset_split_input = gr.Dropdown(['default'], value=['default'], label='Dataset Split', visible=False)
+            dataset_config_input = gr.Dropdown(['default'], value='default', label='Dataset Config', visible=False)
+            dataset_split_input = gr.Dropdown(['default'], value='default', label='Dataset Split', visible=False)
 
             dataset_id_input.change(check_dataset_and_get_config, dataset_id_input, dataset_config_input)
             dataset_config_input.change(

text_classification.py

@@ -100,6 +100,9 @@ def text_classification_fix_column_mapping(column_mapping, ppl, d_id, config, sp
 
     # Infer labels
     id2label_mapping, dataset_labels = text_classification_map_model_and_dataset_labels(id2label, dataset_features)
+    id2label_mapping_dataset_model = {
+        v: k for k, v in id2label_mapping.items()
+    }
     if "data" in column_mapping.keys():
         if isinstance(column_mapping["data"], list):
             # Use the column mapping passed by user
@@ -116,13 +119,13 @@ def text_classification_fix_column_mapping(column_mapping, ppl, d_id, config, sp
     }
     id2label_df = pd.DataFrame({
         "Dataset Labels": dataset_labels,
-        "Model Prediction Labels": [id2label_mapping[label] for label in dataset_labels],
+        "Model Prediction Labels": [id2label_mapping_dataset_model[label] for label in dataset_labels],
     })
 
     if "data" not in column_mapping.keys():
         # Column mapping should contain original model labels
         column_mapping["label"] = {
-            str(i): id2label_mapping[label] for i, label in zip(id2label.keys(), dataset_labels)
+            str(i): id2label_mapping_dataset_model[label] for i, label in zip(id2label.keys(), dataset_labels)
         }
 
     return column_mapping, prediction_input, prediction_result, id2label_df