Update space

app.py

@@ -9,7 +9,7 @@ client = InferenceClient("meta-llama/Llama-3.2-3B-Instruct")
 # Global data store for the table
 data = []
 
-def respond(message, history, system_message, max_tokens=2048, temperature=0.7, top_p=0.9):
+def respond(message, history, system_message):
     messages = [{"role": "system", "content": system_message}]
     for val in history:
         if val[0]:
@@ -17,13 +17,14 @@ def respond(message, history, system_message, max_tokens=2048, temperature=0.7,
         if val[1]:
             messages.append({"role": "assistant", "content": val[1]})
     messages.append({"role": "user", "content": message})
+    
     response = ""
     for message in client.chat_completion(
         messages,
-        max_tokens=max_tokens,
+        max_tokens=2048,
         stream=True,
-        temperature=temperature,
-        top_p=top_p,
+        temperature=0.7,
+        top_p=0.9,
     ):
         token = message.choices[0].delta.content
         response += token
@@ -75,10 +76,21 @@ def extract_table(url):
     except Exception as e:
         return f"<p>Error: {str(e)}</p>"
 
-def user_message(message, history):
-    history = history or []
-    history.append((message, None))
-    return "", history
+def add_text(history, text):
+    history = history + [(text, None)]
+    return history
+
+def bot(history, system_message):
+    if not history:
+        return history
+    
+    user_message = history[-1][0]
+    response = ""
+    
+    for chunk in respond(user_message, history[:-1], system_message):
+        response = chunk
+        history[-1] = (user_message, response)
+        yield history
 
 def handle_prepare(index, history):
     try:
@@ -86,9 +98,7 @@ def handle_prepare(index, history):
         if 0 <= index < len(data):
             topic = data[index]["Topic"]
             message = f"Prepare a 10-minute reading for the topic: {topic}"
-            history = history or []
-            history.append((message, None))
-            return history
+            return add_text(history or [], message)
     except:
         return history or []
 
@@ -131,24 +141,27 @@ with gr.Blocks() as demo:
     
     # Submit button handler
     submit.click(
-        fn=user_message,
-        inputs=[msg, chatbot],
-        outputs=[msg, chatbot]
+        fn=add_text,
+        inputs=[chatbot, msg],
+        outputs=chatbot,
+    ).then(
+        fn=lambda: "",
+        outputs=msg,
     ).then(
-        fn=respond,
-        inputs=[msg, chatbot, system_message],
-        outputs=[chatbot]
+        fn=bot,
+        inputs=[chatbot, system_message],
+        outputs=chatbot,
     )
     
     # Prepare button handler
     prepare_button.click(
         fn=handle_prepare,
         inputs=[prepare_topic, chatbot],
-        outputs=[chatbot]
+        outputs=chatbot,
     ).then(
-        fn=respond,
-        inputs=[prepare_topic, chatbot, system_message],
-        outputs=[chatbot]
+        fn=bot,
+        inputs=[chatbot, system_message],
+        outputs=chatbot,
     )
     
     # Clear button handler