[`TokenizerFast`] `can_save_slow_tokenizer` as a property for when...

[`TokenizerFast`] `can_save_slow_tokenizer` as a property for when `vocab_file`'s folder was removed (#25626)

* pad token should be None by default

* fix tests

* nits

* check if isfile vocabfile

* add warning if sp model folder was deleted

* save SPM when missing folder for sloz

* update the ` can_save_slow_tokenizer`  to be a property

* first batch

* second batch

* missing one

src/transformers/models/xglm/tokenization_xglm_fast.py

@@ -134,7 +134,10 @@ class XGLMTokenizerFast(PreTrainedTokenizerFast):
         )
 
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
+
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
 
     def build_inputs_with_special_tokens(
         self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None

src/transformers/models/xlm_prophetnet/tokenization_xlm_prophetnet.py

@@ -189,6 +189,10 @@ class XLMProphetNetTokenizer(PreTrainedTokenizer):
         for k in self.fairseq_tokens_to_ids.keys():
             self.unique_no_split_tokens.append(k)
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
+
     def __getstate__(self):
         state = self.__dict__.copy()
         state["sp_model"] = None

src/transformers/models/xlm_roberta/tokenization_xlm_roberta_fast.py

@@ -166,7 +166,10 @@ class XLMRobertaTokenizerFast(PreTrainedTokenizerFast):
         )
 
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
+
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
 
     def build_inputs_with_special_tokens(
         self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None

src/transformers/models/xlnet/tokenization_xlnet_fast.py

@@ -169,7 +169,10 @@ class XLNetTokenizerFast(PreTrainedTokenizerFast):
         self.remove_space = remove_space
         self.keep_accents = keep_accents
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
+
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
 
     def build_inputs_with_special_tokens(
         self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None

src/transformers/tokenization_utils_fast.py

@@ -90,7 +90,6 @@ class PreTrainedTokenizerFast(PreTrainedTokenizerBase):
 
     vocab_files_names = VOCAB_FILES_NAMES
     slow_tokenizer_class: PreTrainedTokenizer = None
-    can_save_slow_tokenizer: bool = True
 
     def __init__(self, *args, **kwargs):
         tokenizer_object = kwargs.pop("tokenizer_object", None)
@@ -159,6 +158,14 @@ class PreTrainedTokenizerFast(PreTrainedTokenizerBase):
     def is_fast(self) -> bool:
         return True
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        """
+        `bool`: Whether or not the slow tokenizer can be saved. Usually for sentencepiece based slow tokenizer, this
+        can only be `True` if the original `"sentencepiece.model"` was not deleted.
+        """
+        return True
+
     @property
     def vocab_size(self) -> int:
         """