Update app.py

app.py

@@ -3,9 +3,10 @@ from huggingface_hub.utils import HfHubHTTPError
 
 def predict(message, history):
     try:
-        response = gr.Interface.load(
-            "models/meta-llama/Meta-Llama-3.1-8B"
-        ).predict(message, history)
+        # Load the model and create a prediction function
+        iface = gr.Interface.load("models/meta-llama/Meta-Llama-3.1-8B")
+        fn = iface.launch(share=False,prevent_thread_lock=True)  
+        response = fn(message)  # Get the prediction using the function
         history.append((message, response))
         return "", history
     except HfHubHTTPError as e: