Place BigBirdTokenizer in sentencepiece-only objects (#12975)

src/transformers/__init__.py

@@ -157,7 +157,7 @@ _import_structure = {
     "models.bert_generation": ["BertGenerationConfig"],
     "models.bert_japanese": ["BertJapaneseTokenizer", "CharacterTokenizer", "MecabTokenizer"],
     "models.bertweet": ["BertweetTokenizer"],
-    "models.big_bird": ["BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP", "BigBirdConfig", "BigBirdTokenizer"],
+    "models.big_bird": ["BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP", "BigBirdConfig"],
     "models.bigbird_pegasus": [
         "BIGBIRD_PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "BigBirdPegasusConfig",
@@ -308,6 +308,7 @@ if is_sentencepiece_available():
     _import_structure["models.albert"].append("AlbertTokenizer")
     _import_structure["models.barthez"].append("BarthezTokenizer")
     _import_structure["models.bert_generation"].append("BertGenerationTokenizer")
+    _import_structure["models.big_bird"].append("BigBirdTokenizer")
     _import_structure["models.camembert"].append("CamembertTokenizer")
     _import_structure["models.deberta_v2"].append("DebertaV2Tokenizer")
     _import_structure["models.m2m_100"].append("M2M100Tokenizer")
@@ -1822,7 +1823,7 @@ if TYPE_CHECKING:
     from .models.bert_generation import BertGenerationConfig
     from .models.bert_japanese import BertJapaneseTokenizer, CharacterTokenizer, MecabTokenizer
     from .models.bertweet import BertweetTokenizer
-    from .models.big_bird import BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP, BigBirdConfig, BigBirdTokenizer
+    from .models.big_bird import BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP, BigBirdConfig
     from .models.bigbird_pegasus import BIGBIRD_PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP, BigBirdPegasusConfig
     from .models.blenderbot import BLENDERBOT_PRETRAINED_CONFIG_ARCHIVE_MAP, BlenderbotConfig, BlenderbotTokenizer
     from .models.blenderbot_small import (
@@ -1971,6 +1972,7 @@ if TYPE_CHECKING:
         from .models.albert import AlbertTokenizer
         from .models.barthez import BarthezTokenizer
         from .models.bert_generation import BertGenerationTokenizer
+        from .models.big_bird import BigBirdTokenizer
         from .models.camembert import CamembertTokenizer
         from .models.deberta_v2 import DebertaV2Tokenizer
         from .models.m2m_100 import M2M100Tokenizer

src/transformers/utils/dummy_sentencepiece_objects.py

@@ -29,6 +29,15 @@ class BertGenerationTokenizer:
         requires_backends(cls, ["sentencepiece"])
 
 
+class BigBirdTokenizer:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["sentencepiece"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["sentencepiece"])
+
+
 class CamembertTokenizer:
     def __init__(self, *args, **kwargs):
         requires_backends(self, ["sentencepiece"])