Add Mirror Option for Downloads (#6679)

* Add Tuna Mirror for Downloads from China

* format fix

* Use preset instead of hardcoding URL

* Fix

* make style

* update the mirror option doc

* update the mirror

src/transformers/file_utils.py

@@ -141,6 +141,10 @@ DUMMY_MASK = [[1, 1, 1, 1, 1], [1, 1, 1, 0, 0], [0, 0, 0, 1, 1]]
 
 S3_BUCKET_PREFIX = "https://s3.amazonaws.com/models.huggingface.co/bert"
 CLOUDFRONT_DISTRIB_PREFIX = "https://cdn.huggingface.co"
+PRESET_MIRROR_DICT = {
+    "tuna": "https://mirrors.tuna.tsinghua.edu.cn/hugging-face-models",
+    "bfsu": "https://mirrors.bfsu.edu.cn/hugging-face-models",
+}
 
 
 def is_torch_available():
@@ -570,7 +574,7 @@ def is_remote_url(url_or_filename):
     return parsed.scheme in ("http", "https")
 
 
-def hf_bucket_url(model_id: str, filename: str, use_cdn=True) -> str:
+def hf_bucket_url(model_id: str, filename: str, use_cdn=True, mirror=None) -> str:
     """
     Resolve a model identifier, and a file name, to a HF-hosted url
     on either S3 or Cloudfront (a Content Delivery Network, or CDN).
@@ -586,7 +590,13 @@ def hf_bucket_url(model_id: str, filename: str, use_cdn=True) -> str:
     are not shared between the two because the cached file's name contains
     a hash of the url.
     """
-    endpoint = CLOUDFRONT_DISTRIB_PREFIX if use_cdn else S3_BUCKET_PREFIX
+    endpoint = (
+        PRESET_MIRROR_DICT.get(mirror, mirror)
+        if mirror
+        else CLOUDFRONT_DISTRIB_PREFIX
+        if use_cdn
+        else S3_BUCKET_PREFIX
+    )
     legacy_format = "/" not in model_id
     if legacy_format:
         return f"{endpoint}/{model_id}-{filename}"