Update app.py

app.py

@@ -66,6 +66,15 @@ def chat_accordion():
             interactive=True,
             label="p (nucleus sampling)",
         )
+        
+        max_tokens = gr.Slider(
+            minimum=64,
+            maximum=1024,
+            value=64,
+            step=1,
+            interactive=True,
+            label="Max Tokens",
+        )
 
         session_id = gr.Textbox(
             value=uuid4,
@@ -105,7 +114,7 @@ def chat_accordion():
                     visible=False,
                 )
 
-    return temperature, top_p, instructions, user_name, bot_name, session_id
+    return temperature, top_p, instructions, user_name, bot_name, session_id, max_tokens
 
 
 def format_chat_prompt(
@@ -162,6 +171,7 @@ def chat_tab():
         temperature: float,
         top_p: float,
         session_id: str,
+        max_tokens: int
     ):
         prompt = format_chat_prompt(message, history, instructions, user_name, bot_name)
         payload = {
@@ -169,7 +179,7 @@ def chat_tab():
             "data": {
                 "inputs": prompt,
                 "parameters": {
-                    "max_new_tokens": 512,
+                    "max_new_tokens": max_tokens,
                     "do_sample": True,
                     "top_p": top_p,
                     "stop": ["User:"],
@@ -237,6 +247,7 @@ def chat_tab():
                 user_name,
                 bot_name,
                 session_id,
+                max_tokens
             ) = chat_accordion()
 
         with gr.Column():
@@ -271,6 +282,7 @@ def chat_tab():
                         temperature,
                         top_p,
                         session_id,
+                        max_tokens
                     ],
                     submit_btn="Send",
                     stop_btn="Stop",