[TokenizerFast] can_save_slow_tokenizer as a property for when `v…

…ocab_file`'s folder was removed (huggingface#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/albert/tokenization_albert_fast.py

@@ -165,7 +165,10 @@ def __init__(
         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/models/barthez/tokenization_barthez_fast.py

@@ -146,7 +146,10 @@ def __init__(
         )
 
         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/bertweet/tokenization_bertweet.py

@@ -398,8 +398,12 @@ def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] =
             save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["merges_file"]
         )
 
-        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
             copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
 
         if os.path.abspath(self.merges_file) != os.path.abspath(out_merge_file):
             copyfile(self.merges_file, out_merge_file)

src/transformers/models/big_bird/tokenization_big_bird_fast.py

@@ -150,7 +150,10 @@ def __init__(
         )
 
         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/camembert/tokenization_camembert_fast.py

@@ -140,7 +140,10 @@ def __init__(
         )
 
         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/cpm/tokenization_cpm_fast.py

@@ -141,7 +141,6 @@ def __init__(
         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
 
         try:
             import jieba
@@ -153,6 +152,10 @@ def __init__(
         self.jieba = jieba
         self.translator = str.maketrans(" \n", "\u2582\u2583")
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
+
     # Copied from transformers.models.xlnet.tokenization_xlnet_fast.XLNetTokenizerFast.build_inputs_with_special_tokens
     def build_inputs_with_special_tokens(
         self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None

src/transformers/models/deberta_v2/tokenization_deberta_v2_fast.py

@@ -148,7 +148,10 @@ def __init__(
         self.do_lower_case = do_lower_case
         self.split_by_punct = split_by_punct
         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, token_ids_1=None):
         """

src/transformers/models/fnet/tokenization_fnet_fast.py

@@ -132,7 +132,10 @@ def __init__(
         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/models/layoutxlm/tokenization_layoutxlm_fast.py

@@ -257,7 +257,6 @@ def __init__(
         )
 
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
 
         # additional properties
         self.cls_token_box = cls_token_box
@@ -266,6 +265,10 @@ def __init__(
         self.pad_token_label = pad_token_label
         self.only_label_first_subword = only_label_first_subword
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
+
     @add_end_docstrings(LAYOUTXLM_ENCODE_KWARGS_DOCSTRING)
     def __call__(
         self,

src/transformers/models/llama/tokenization_llama_fast.py

@@ -128,7 +128,10 @@ def __init__(
         self.update_post_processor()
         self.use_default_system_prompt = use_default_system_prompt
         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 update_post_processor(self):
         """

src/transformers/models/mbart/tokenization_mbart_fast.py

@@ -129,7 +129,6 @@ def __init__(
         )
 
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
 
         _additional_special_tokens = FAIRSEQ_LANGUAGE_CODES.copy()
 
@@ -149,6 +148,10 @@ def __init__(
         self.tgt_lang = tgt_lang
         self.set_src_lang_special_tokens(self._src_lang)
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
+
     @property
     def src_lang(self) -> str:
         return self._src_lang

src/transformers/models/mbart50/tokenization_mbart50_fast.py

@@ -147,7 +147,6 @@ def __init__(
         )
 
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
 
         self.lang_code_to_id = {
             lang_code: self.convert_tokens_to_ids(lang_code) for lang_code in FAIRSEQ_LANGUAGE_CODES
@@ -158,6 +157,10 @@ def __init__(
         self.cur_lang_code_id = self.lang_code_to_id[self._src_lang]
         self.set_src_lang_special_tokens(self._src_lang)
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
+
     @property
     def src_lang(self) -> str:
         return self._src_lang

src/transformers/models/mluke/tokenization_mluke.py

@@ -1494,8 +1494,12 @@ def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] =
             save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
         )
 
-        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
             copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
 
         entity_vocab_file = os.path.join(
             save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["entity_vocab_file"]

src/transformers/models/nllb/tokenization_nllb_fast.py

@@ -175,7 +175,6 @@ def __init__(
         )
 
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
 
         _additional_special_tokens = FAIRSEQ_LANGUAGE_CODES.copy()
 
@@ -195,6 +194,10 @@ def __init__(
         self.tgt_lang = tgt_lang
         self.set_src_lang_special_tokens(self._src_lang)
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
+
     @property
     def src_lang(self) -> str:
         return self._src_lang

src/transformers/models/pegasus/tokenization_pegasus_fast.py

@@ -152,7 +152,10 @@ def __init__(
             **kwargs,
         )
         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 _special_token_mask(self, seq):
         all_special_ids = set(self.all_special_ids)  # call it once instead of inside list comp

src/transformers/models/phobert/tokenization_phobert.py

@@ -324,8 +324,12 @@ def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] =
             save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["merges_file"]
         )
 
-        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
             copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
 
         if os.path.abspath(self.merges_file) != os.path.abspath(out_merge_file):
             copyfile(self.merges_file, out_merge_file)

src/transformers/models/reformer/tokenization_reformer_fast.py

@@ -110,7 +110,10 @@ def __init__(
         )
 
         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 save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
         if not self.can_save_slow_tokenizer:

src/transformers/models/rembert/tokenization_rembert.py

@@ -263,7 +263,11 @@ def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] =
             save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
         )
 
-        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
             copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
 
         return (out_vocab_file,)

src/transformers/models/rembert/tokenization_rembert_fast.py

@@ -139,7 +139,10 @@ def __init__(
         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/models/t5/tokenization_t5_fast.py

@@ -142,9 +142,12 @@ def __init__(
         )
 
         self.vocab_file = vocab_file
-        self.can_save_slow_tokenizer = False if not self.vocab_file else True
         self._extra_ids = extra_ids
 
+    @property
+    def can_save_slow_tokenizer(self) -> bool:
+        return os.path.isfile(self.vocab_file) if self.vocab_file else False
+
     @staticmethod
     def _eventually_correct_t5_max_length(pretrained_model_name_or_path, max_model_length, init_max_model_length):
         if pretrained_model_name_or_path in T5TokenizerFast.max_model_input_sizes:

src/transformers/models/xglm/tokenization_xglm_fast.py

@@ -134,7 +134,10 @@ def __init__(
         )
 
         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 @@ def __init__(
         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 @@ def __init__(
         )
 
         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 @@ def __init__(
         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 @@ def __init__(self, *args, **kwargs):
     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:
         """