add cuda

app.py

@@ -25,7 +25,7 @@ logger = logging.getLogger(__name__)
 login(token=os.environ['HF_TOKEN'])
 
 # Global variables
-CUR_MODEL = "./repllama-instruct-hard-positives-v2-joint"
+CUR_MODEL = "orionweller/repllama-instruct-hard-positives-v2-joint"
 BASE_MODEL = "meta-llama/Llama-2-7b-hf"
 tokenizer = None
 model = None