Update app.py

app.py

@@ -36,7 +36,7 @@ def start_worker(model_path: str, bits=16):
         "-m",
         "llava.serve.model_worker",
         "--host",
-        "ec2-3-239-127-158.compute-1.amazonaws.com",
+        "0.0.0.0",
         "--controller",
         "http://localhost:10000",
         "--model-path",
@@ -53,7 +53,7 @@ def start_worker(model_path: str, bits=16):
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument("--host", type=str, default="ec2-3-239-127-158.compute-1.amazonaws.com")
+    parser.add_argument("--host", type=str, default="0.0.0.0")
     parser.add_argument("--port", type=int)
     parser.add_argument("--controller-url", type=str, default="http://localhost:10000")
     parser.add_argument("--concurrency-count", type=int, default=5)
@@ -78,7 +78,7 @@ Set the environment variable `model` to change the model:
     print(f"args: {gws.args}")
 
     model_path = os.getenv("model", "liuhaotian/llava-v1.6-mistral-7b")
-    bits = int(os.getenv("bits", 8))
+    bits = int(os.getenv("bits", 4))
     concurrency_count = int(os.getenv("concurrency_count", 5))
 
     controller_proc = start_controller()
@@ -95,7 +95,7 @@ Set the environment variable `model` to change the model:
             api_open=False
         ).launch(
             server_name=gws.args.host,
-            server_port=8080,
+            server_port=gws.args.port,
             share=True
         )
 
@@ -106,4 +106,4 @@ Set the environment variable `model` to change the model:
         worker_proc.kill()
         controller_proc.kill()
 
-        sys.exit(exit_status)
+        sys.exit(exit_status)