Update app.py

app.py

@@ -107,6 +107,7 @@ def summarize(text):
         # top_p=0.9,
         # do_sample=True,
         # temperature=0.6,
+        min_new_tokens=50,
         max_new_tokens=2048,
         logits_processor=[xgr_logits_processor]
     )
@@ -116,7 +117,7 @@ def summarize(text):
     ]
     
     response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
-    return response
+    return response[0]
 
 with gr.Blocks() as demo:
     gr.Markdown("## 📝 Structured Information Extraction for Clinical Reports")
@@ -174,7 +175,7 @@ with gr.Blocks() as demo:
             fn=summarize,
             inputs=input_text,
             outputs=output_text,
-            cache_examples="lazy",
+            # cache_examples="lazy",
         )
 
     summarize_btn.click(