diff --git a/cnocr/consts.py b/cnocr/consts.py
index b3c2741..a87dfc8 100644
--- a/cnocr/consts.py
+++ b/cnocr/consts.py
@@ -1,5 +1,5 @@
 # coding: utf-8
-# Copyright (C) 2021, [Breezedeus](https://github.com/breezedeus).
+# Copyright (C) 2021-2023, [Breezedeus](https://github.com/breezedeus).
 # Licensed to the Apache Software Foundation (ASF) under one
 # or more contributor license agreements.  See the NOTICE file
 # distributed with this work for additional information
@@ -17,6 +17,7 @@
 # specific language governing permissions and limitations
 # under the License.
 
+import os
 import string
 from collections import OrderedDict
 from pathlib import Path
@@ -32,6 +33,7 @@
 # 模型版本只对应到第二层，第三层的改动表示模型兼容。
 # 如: __version__ = '2.2.*'，对应的 MODEL_VERSION 都是 '2.2'
 MODEL_VERSION = '.'.join(__version__.split('.', maxsplit=2)[:2])
+DOWNLOAD_SOURCE = os.environ.get('CNOCR_DOWNLOAD_SOURCE', 'CN')
 
 IMG_STANDARD_HEIGHT = 32
 CN_VOCAB_FP = Path(__file__).parent.absolute() / 'label_cn.txt'
@@ -135,20 +137,25 @@
 HF_HUB_SUBFOLDER = "models/cnocr/%s" % MODEL_VERSION
 PAID_HF_HUB_REPO_ID = "breezedeus/paid-models"
 PAID_HF_HUB_SUBFOLDER = "cnocr/%s" % MODEL_VERSION
+CN_OSS_ENDPOINT = (
+    "https://sg-models.oss-cn-beijing.aliyuncs.com/cnocr/%s/" % MODEL_VERSION
+)
 
 
 def format_hf_hub_url(url: str, is_paid_model=False) -> dict:
+    out_dict = {'filename': url}
+
     if is_paid_model:
         repo_id = PAID_HF_HUB_REPO_ID
         subfolder = PAID_HF_HUB_SUBFOLDER
     else:
         repo_id = HF_HUB_REPO_ID
         subfolder = HF_HUB_SUBFOLDER
-    return {
-        'repo_id': repo_id,
-        'subfolder': subfolder,
-        'filename': url,
-    }
+        out_dict['cn_oss'] = CN_OSS_ENDPOINT
+    out_dict.update(
+        {'repo_id': repo_id, 'subfolder': subfolder,}
+    )
+    return out_dict
 
 
 class AvailableModels(object):
diff --git a/cnocr/ppocr/pp_recognizer.py b/cnocr/ppocr/pp_recognizer.py
index d535017..8d97528 100755
--- a/cnocr/ppocr/pp_recognizer.py
+++ b/cnocr/ppocr/pp_recognizer.py
@@ -33,7 +33,7 @@
 from .postprocess import build_post_process
 from .utility import create_predictor
 from .consts import PP_SPACE
-from ..consts import MODEL_VERSION, AVAILABLE_MODELS
+from ..consts import MODEL_VERSION, AVAILABLE_MODELS, DOWNLOAD_SOURCE
 
 
 logger = logging.getLogger(__name__)
@@ -106,7 +106,7 @@ def _assert_and_prepare_model_files(self, model_fp, root):
                     % ((self._model_name, self._model_backend),)
                 )
             url = AVAILABLE_MODELS.get_url(self._model_name, self._model_backend)
-            get_model_file(url, self._model_dir)
+            get_model_file(url, self._model_dir, download_source=DOWNLOAD_SOURCE)  # download the .zip file and unzip
 
         self._model_fp = model_fp
         logger.info('use model: %s' % self._model_fp)
diff --git a/cnocr/recognizer.py b/cnocr/recognizer.py
index 17fe7a8..b97330c 100644
--- a/cnocr/recognizer.py
+++ b/cnocr/recognizer.py
@@ -28,7 +28,7 @@
 import torch
 from cnstd.utils import get_model_file
 
-from .consts import MODEL_VERSION, AVAILABLE_MODELS
+from .consts import MODEL_VERSION, AVAILABLE_MODELS, DOWNLOAD_SOURCE
 from .models.ocr_model import OcrModel
 from .utils import (
     data_dir,
@@ -177,7 +177,7 @@ def _assert_and_prepare_model_files(self, model_fp, root):
                     % ((self._model_name, self._model_backend),)
                 )
             url = AVAILABLE_MODELS.get_url(self._model_name, self._model_backend)
-            get_model_file(url, self._model_dir)
+            get_model_file(url, self._model_dir, download_source=DOWNLOAD_SOURCE)  # download the .zip file and unzip
             fps = glob(
                 '%s/%s*.%s' % (self._model_dir, self._model_file_prefix, model_ext)
             )