return text

app.py

@@ -87,7 +87,7 @@ def _prep_data_for_input(image):
         text=prompt
     )
 
-@spaces.GPU
+@spaces.GPU(duration=120)
 def generate_response(image):
     inputs = _prep_data_for_input(image)
     inputs = {k: v.to(model.device).unsqueeze(0) for k, v in inputs.items()}
@@ -100,10 +100,10 @@ def generate_response(image):
     output_text = processor.tokenizer.decode(generated_tokens, skip_special_tokens=True)
 
     try:
-        return json.loads(output_text)
+        return str(json.loads(output_text))
     except Exception:
         gr.Warning("Failed to parse JSON from output")
-        return {}
+        return output_text
 
 title = "ColPali fine-tuning Query Generator"
 description = """[ColPali](https://huggingface.co/papers/2407.01449) is a very exciting new approach to multimodal document retrieval which aims to replace existing document retrievers which often rely on an OCR step with an end-to-end multimodal approach. 
@@ -129,7 +129,7 @@ examples = [
 demo = gr.Interface(
     fn=generate_response,
     inputs=gr.Image(type="pil"),
-    outputs=gr.Json(),
+    outputs=gr.Text(),
     title=title,
     description=description,
     examples=examples,