[XGLMTokenizer] fix init and add in AutoTokenizer (#15406)

src/transformers/models/auto/tokenization_auto.py

@@ -225,6 +225,13 @@ else:
                     None,
                 ),
             ),
+            (
+                "xglm",
+                (
+                    "XGLMTokenizer" if is_sentencepiece_available() else None,
+                    "XGLMTokenizerFast" if is_tokenizers_available() else None,
+                ),
+            ),
         ]
     )
 

src/transformers/models/xglm/__init__.py

@@ -18,14 +18,22 @@
 from typing import TYPE_CHECKING
 
 # rely on isort to merge the imports
-from ...file_utils import _LazyModule, is_flax_available, is_tokenizers_available, is_torch_available
+from ...file_utils import (
+    _LazyModule,
+    is_flax_available,
+    is_sentencepiece_available,
+    is_tokenizers_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
     "configuration_xglm": ["XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP", "XGLMConfig"],
-    "tokenization_xglm": ["XGLMTokenizer"],
 }
 
+if is_sentencepiece_available():
+    _import_structure["tokenization_xglm"] = ["XGLMTokenizer"]
+
 if is_tokenizers_available():
     _import_structure["tokenization_xglm_fast"] = ["XGLMTokenizerFast"]
 
@@ -48,7 +56,9 @@ if is_flax_available():
 
 if TYPE_CHECKING:
     from .configuration_xglm import XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XGLMConfig
-    from .tokenization_xglm import XGLMTokenizer
+
+    if is_sentencepiece_available():
+        from .tokenization_xglm import XGLMTokenizer
 
     if is_tokenizers_available():
         from .tokenization_xglm_fast import XGLMTokenizerFast