Update app.py

app.py

@@ -5,22 +5,31 @@ model_name = "facebook/blenderbot-400M-distill"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
 
-history = []
+history = []  # list of (user_message, bot_response) pairs
 
-def respond(message):
-    global history
-    history.append(message)
-    input_text = " ".join(history[-5:])  # keep last 5 messages
+def respond(message, chat_history):
+    # chat_history is a list of tuples: (user, bot)
+    if chat_history is None:
+        chat_history = []
+    # add user message to history
+    chat_history.append((message, None))
+
+    # flatten last few messages for context, e.g. last 5 user+bot turns (10 messages)
+    # but keep it simple here and just join user messages
+    input_text = " ".join([msg for pair in chat_history[-5:] for msg in pair if msg])
     inputs = tokenizer(input_text, return_tensors="pt")
     outputs = model.generate(**inputs, max_new_tokens=100)
     response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    history.append(response)
-    return response
+
+    # update last user message's bot response
+    chat_history[-1] = (message, response)
+
+    return chat_history, chat_history
 
 with gr.Blocks() as demo:
     chatbot = gr.Chatbot()
     msg = gr.Textbox(placeholder="Say something...")
-    msg.submit(respond, msg, chatbot)
+    msg.submit(respond, inputs=[msg, chatbot], outputs=[chatbot, chatbot])
 
 demo.launch(css="""
 body {background-color: #000 !important; color: #fff !important;}