Add top-p parameter to respond function and update ChatInterface

app.py

@@ -11,6 +11,7 @@ def respond(
     system_message: str,
     max_new_tokens: int,
     temperature: float,
+    top_p: float,
 ) -> str:
     """
     Respond to user input using the model.
@@ -18,6 +19,7 @@ def respond(
     # Set the model parameters
     model.temperature = temperature
     model.max_output_tokens = max_new_tokens
+    model.top_p = top_p
 
     history_langchain_format = []
     # Add the dialog history to the history
@@ -31,7 +33,7 @@ def respond(
     # Combine the system message, history, and user input into a single list
     model_input = [SystemMessage(content=system_message)] + \
         history_langchain_format + [HumanMessage(content=user_input)]
-    
+
     response = model.invoke(model_input)
     return response.content
 
@@ -49,6 +51,13 @@ demo = gr.ChatInterface(
                   step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7,
                   step=0.1, label="Temperature"),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=0.95,
+            step=0.05,
+            label="Top-p (nucleus sampling)",
+        ),
     ],
 )