GSK-2909-handle-input-too-long-error (#165)

- fix labels not found; handle input too long (6457b7ab3e357b83e70bad7268307a81d2154bfa)
- Remove leading spaces (b709c217e8e05020af6b864792c2941358e29dfc)
- Remove tailing spaces (6dbdec2b3e7c630a187b0171e19325eb4e2f1126)
- Remove leading spaces (c1a5e7ecbc5505f334a5bf1a458bf1d7ffaa00e3)


Co-authored-by: zcy <[email protected]>

app_text_classification.py

@@ -201,6 +201,7 @@ def get_demo():
     gr.on(
         triggers=[
             model_id_input.change,
+            model_id_input.input,
             dataset_id_input.change,
             dataset_config_input.change,
             dataset_split_input.change,

text_classification.py

@@ -28,10 +28,14 @@ def get_labels_and_features_from_dataset(ds):
         if len(label_keys) == 0: # no labels found
             # return everything for post processing
             return list(dataset_features.keys()), list(dataset_features.keys()), None
+
+        labels = None
         if not isinstance(dataset_features[label_keys[0]], datasets.ClassLabel):
             if hasattr(dataset_features[label_keys[0]], "feature"):
                 label_feat = dataset_features[label_keys[0]].feature
                 labels = label_feat.names
+            else:
+                labels = ds.unique(label_keys[0])
         else:
             labels = dataset_features[label_keys[0]].names
         return labels, features, label_keys
@@ -83,9 +87,17 @@ def hf_inference_api(model_id, hf_token, payload):
     url = f"{hf_inference_api_endpoint}/models/{model_id}"
     headers = {"Authorization": f"Bearer {hf_token}"}
     response = requests.post(url, headers=headers, json=payload)
+
     if not hasattr(response, "status_code") or response.status_code != 200:
         logger.warning(f"Request to inference API returns {response}")
+
     try:
+        output = response.json()
+        if "error" in output and "Input is too long" in output["error"]:
+          payload.update({"parameters": {"truncation": True, "max_length": 512}})
+          response = requests.post(url, headers=headers, json=payload)
+          if not hasattr(response, "status_code") or response.status_code != 200:
+              logger.warning(f"Request to inference API returns {response}")
         return response.json()
     except Exception:
         return {"error": response.content}

text_classification_ui_helpers.py

@@ -341,8 +341,8 @@ def align_columns_and_show_prediction(
     ):
         return (
             gr.update(value=MAPPING_STYLED_ERROR_WARNING, visible=True),
-            gr.update(visible=False),
-            gr.update(visible=False),
+            gr.update(value=prediction_input, lines=min(len(prediction_input)//225 + 1, 5), visible=True),
+            gr.update(value=prediction_response, visible=True),
             gr.update(visible=True, open=True),
             gr.update(interactive=(run_inference and inference_token != "")),
             "",
@@ -351,7 +351,7 @@ def align_columns_and_show_prediction(
 
     return (
         gr.update(value=VALIDATED_MODEL_DATASET_STYLED, visible=True),
-        gr.update(value=prediction_input, lines=len(prediction_input)//225 + 1, visible=True),
+        gr.update(value=prediction_input, lines=min(len(prediction_input)//225 + 1, 5), visible=True),
         gr.update(value=prediction_response, visible=True),
         gr.update(visible=True, open=False),
         gr.update(interactive=(run_inference and inference_token != "")),