Refactor

app.py

@@ -346,12 +346,7 @@ with st.expander("Advanced configuration"):
     )
 
 with st.form(key="form"):
-    # Grab all models fine-tuned on SQuAD for question answering tasks
-    if selected_task == "extractive_question_answering":
-        compatible_models = get_compatible_models(selected_task, [selected_dataset, "squad", "squad_v2"])
-    else:
-        compatible_models = get_compatible_models(selected_task, [selected_dataset])
-
+    compatible_models = get_compatible_models(selected_task, [selected_dataset])
     selected_models = st.multiselect(
         "Select the models you wish to evaluate",
         compatible_models,

evaluation.py

@@ -43,7 +43,7 @@ def filter_evaluated_models(models, task, dataset_name, dataset_config, dataset_
         )
         candidate_id = hash(evaluation_info)
         if candidate_id in evaluation_ids:
-            st.info(f"Model {model} has already been evaluated on this configuration. Skipping evaluation...")
+            st.info(f"Model `{model}` has already been evaluated on this configuration. Skipping evaluation...")
             models.pop(idx)
 
     return models

utils.py

@@ -75,6 +75,9 @@ def get_compatible_models(task: str, dataset_ids: List[str]) -> List[str]:
     """
     # TODO: relax filter on PyTorch models if TensorFlow supported in AutoTrain
     compatible_models = []
+    if task == "extractive_question_answering":
+        dataset_ids.extend(["squad", "squad_v2"])
+
     for dataset_id in dataset_ids:
         model_filter = ModelFilter(
             task=AUTOTRAIN_TASK_TO_HUB_TASK[task],
@@ -82,7 +85,7 @@ def get_compatible_models(task: str, dataset_ids: List[str]) -> List[str]:
             library=["transformers", "pytorch"],
         )
         compatible_models.extend(HfApi().list_models(filter=model_filter))
-    return sorted([model.modelId for model in compatible_models])
+    return set(sorted([model.modelId for model in compatible_models]))
 
 
 def get_key(col_mapping, val):