Update

m3_server.py

@@ -15,9 +15,9 @@ from pydantic import BaseModel
 from starlette.status import HTTP_504_GATEWAY_TIMEOUT
 
 Path("/tmp/cache").mkdir(exist_ok=True)
-os.environ[
-    "HF_HOME"
-] = "/tmp/cache"  # does not quite work, need Path("/tmp/cache").mkdir(exist_ok=True)?
+os.environ["HF_HOME"] = "/tmp/cache"
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/cache"
+# does not quite work, need Path("/tmp/cache").mkdir(exist_ok=True)?
 
 batch_size = 2  # gpu batch_size in order of your available vram
 max_request = 10  # max request for future improvements on api calls / gpu batches (for now is pretty basic)
@@ -31,6 +31,13 @@ port = 3000
 port = 7860
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
+os.environ["TZ"] = "Asia/Shanghai"
+try:
+    time.tzset()  # type: ignore # pylint: disable=no-member
+except Exception:
+    # Windows
+    print("Windows, cant run time.tzset()")
+
 
 class m3Wrapper:
     def __init__(self, model_name: str, device: str = DEVICE):