update to bf16 model

app.py

@@ -75,7 +75,7 @@ def generate_response(message, history, initial_prompt, user_role, assistant_rol
         full_prompt = initial_prompt + "\n\n" + formatted_input + f"<{assistant_role}>:"
 
     completion = client.completions.create(
-        model="meta-llama/Meta-Llama-3.1-405B-FP8",
+        model="meta-llama/Meta-Llama-3.1-405B",
         prompt=full_prompt,
         temperature=0.7,
         frequency_penalty=0.1,