vlm: optimize tensor transport (#6003)

Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>

python/sglang/utils.py

@@ -14,6 +14,7 @@ import traceback
 import urllib.request
 import weakref
 from concurrent.futures import ThreadPoolExecutor
+from functools import wraps
 from io import BytesIO
 from json import dumps
 from typing import Any, Callable, List, Optional, Tuple, Type, Union
@@ -28,6 +29,24 @@ from tqdm import tqdm
 logger = logging.getLogger(__name__)
 
 
+def execute_once(func):
+    has_run = None
+
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        nonlocal has_run
+        if not has_run:
+            func(*args, **kwargs)
+            has_run = True
+
+    return wrapper
+
+
+@execute_once
+def info_once(message: str):
+    logger.info(message)
+
+
 def convert_json_schema_to_str(json_schema: Union[dict, str, Type[BaseModel]]) -> str:
     """Convert a JSON schema to a string.
     Parameters