Update main.py

main.py

@@ -37,11 +37,11 @@ def generate(prompt, history, temperature=0.2, max_new_tokens=30000, top_p=0.95,
         seed=42,
     )
     formatted_prompt = format_prompt(prompt, history)
-    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
+    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
 
-    # Accumula l'output in una stringa anziché una lista
-    output_text = ""
+    # Accumula l'output in una lista
+    output_list = []
     for response in stream:
-        output_text += response.token.text
+        output_list.append(response.token.text)
 
-    return output_text  # Restituisci l'intero testo generato come una stringa
+    return iter(output_list)  # Restituisci la lista come un iteratore