Update app.py

app.py

@@ -525,10 +525,9 @@ def get_response_from_llama(query, model, selected_docs, file_type, num_calls=1,
             # Generate content with streaming enabled
             for message in client.chat_completion(
                 messages=messages,
-                max_new_tokens=20000,
+                max_tokens=20000,
                 temperature=temperature,
                 stream=True,
-                top_p=0.8,
             ):
                 if message.choices and message.choices[0].delta and message.choices[0].delta.content:
                     chunk = message.choices[0].delta.content
@@ -776,7 +775,7 @@ def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=
             logging.info(f"API call {i+1}/{num_calls}")
             for message in client.chat_completion(
                 messages=messages,
-                max_new_tokens=20000,
+                max_tokens=20000,
                 temperature=temperature,
                 stream=True,
                 top_p=0.8,