Update app.py

Made it so that the max_new_tokens can be changed!

app.py

@@ -12,7 +12,7 @@ def format_prompt(message, history):
   return prompt
 
 def generate(
-    prompt, history, temperature=0.5, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0,
+    prompt, history, temperature=0.5, max_new_tokens=512, top_p=0.95, repetition_penalty=1.0,
 ):
     temperature = float(temperature)
     if temperature < 1e-2:
@@ -48,6 +48,15 @@ additional_inputs=[
         step=0.05,
         interactive=True,
         info="Higher values generate more diverse outputs",
+    ),
+    gr.Slider(
+        label="max_new_tokens",
+        value=512,
+        minimum=1,
+        maximum=2048,
+        step=1,
+        interactive=True,
+        info="Higher values allows for longer answers",
     )
 ]