.

app.py

@@ -37,9 +37,9 @@ def respond(
     message,
     history,
     system_message,
-    max_tokens= 256,
-    temperature= 0.3,
-    top_p= 0.95,
+    max_tokens=256,
+    temperature=0.3,
+    top_p=0.95,
 ):
     # Construct the chat list
     chat_list = [{"role": "system", "content": system_message}]
@@ -65,10 +65,13 @@ def respond(
     # Adjust max_length to be greater than input_length
     max_length = input_length + max_tokens
 
+    # Ensure max_length is an integer
+    max_length = int(max_length)
+
     outputs = pipe(
         prompt,
-        max_new_tokens=max_tokens,
-        max_length=max_length,  # Add this line
+        max_new_tokens=int(max_tokens),  # Ensure this is an integer
+        max_length=max_length,
         num_beams=1,
         do_sample=True,
         temperature=temperature,
@@ -97,7 +100,7 @@ demo = gr.ChatInterface(
             value="You are General Knowledge Assistant. Answer the questions based on the provided information. Be succinct and use first-principles thinking to answer the questions.",
             label="System message",
         ),
-        gr.Slider(minimum=1, maximum=2048, value=256, step=1, label="Max new tokens"),
+        gr.Slider(minimum=1, maximum=2048, value=256, step=1, label="Max new tokens", dtype=int),
         gr.Slider(minimum=0.1, maximum=2.0, value=0.3, step=0.1, label="Temperature"),
         gr.Slider(
             minimum=0.1,