Update app.py

app.py

@@ -534,6 +534,9 @@ def start_worker(model_path: str):
         "http://localhost:10000",
         "--model-path",
         model_path,
+        "--model-name",
+        "llava-v1.5-13b-4bit",
+        "--load-4bit",
     ]
     return subprocess.Popen(worker_command)
 
@@ -582,7 +585,7 @@ if __name__ == "__main__":
     args = get_args()
     logger.info(f"args: {args}")
 
-    model_path = "liuhaotian/llava-v1.5-7b"
+    model_path = "liuhaotian/llava-v1.5-13b"
 
     preload_models(model_path)