Update app.py

app.py

@@ -9,8 +9,11 @@ access_token = os.getenv("Llama2")
 
 def greet(text):
 
-    model = "meta-llama/Llama-2-7b-hf"
-    tokenizer = AutoTokenizer.from_pretrained(model, token=access_token)
+    model = "TheBloke/Llama-2-7b-Chat-GGUF"
+    tokenizer = AutoTokenizer.from_pretrained(model, 
+                                              token=access_token, 
+                                              # model_file="llama-2-7b-chat.Q2_K.gguf"
+                                             )
     
     pipeline = transformers.pipeline(
     "text-generation",