Update app.py

app.py

@@ -40,7 +40,7 @@ def find_id(prediction_list, inp):
 
 def live_inference(prompt, max_new_tokens = 1024, top_k = 50, top_p = 0.9, temperature = 0.6, presence_penalty=1.15):
   output = replicate.run(
-      "meta/meta-llama-3.1-405b-instruct",
+      "meta/meta-llama-3-8b-instruct",
       input={
           "top_k": top_k,
           "top_p": top_p,