diff --git a/setup.py b/setup.py
index 95aeea6d3a..c43296bf0e 100644
--- a/setup.py
+++ b/setup.py
@@ -174,7 +174,7 @@ _deps = [
     "tf2onnx",
     "timeout-decorator",
     "timm",
-    "tokenizers>=0.14,<0.19",
+    "tokenizers>=0.16,<0.19",
     "torch",
     "torchaudio",
     "torchvision",
diff --git a/src/transformers/convert_slow_tokenizer.py b/src/transformers/convert_slow_tokenizer.py
index 1980ba643a..808ad37db2 100644
--- a/src/transformers/convert_slow_tokenizer.py
+++ b/src/transformers/convert_slow_tokenizer.py
@@ -601,14 +601,15 @@ class SpmConverter(Converter):
         if hasattr(self.original_tokenizer, "legacy") and not self.original_tokenizer.legacy:
             prepend_scheme = "first"
         return pre_tokenizers.Metaspace(
-            replacement=replacement, add_prefix_space=add_prefix_space, prepend_scheme=prepend_scheme
+            replacement=replacement, prepend_scheme=prepend_scheme
         )
 
     def post_processor(self):
         return None
 
     def decoder(self, replacement, add_prefix_space):
-        return decoders.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space)
+        prepend_scheme = "always" if add_prefix_space else "never"
+        return decoders.Metaspace(replacement=replacement, prepend_scheme=prepend_scheme)
 
     def converted(self) -> Tokenizer:
         tokenizer = self.tokenizer(self.proto)
@@ -722,7 +723,8 @@ class DebertaV2Converter(SpmConverter):
         list_pretokenizers = []
         if self.original_tokenizer.split_by_punct:
             list_pretokenizers.append(pre_tokenizers.Punctuation(behavior="isolated"))
-        list_pretokenizers.append(pre_tokenizers.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space))
+        prepend_scheme = "always" if add_prefix_space else "never"
+        list_pretokenizers.append(pre_tokenizers.Metaspace(replacement=replacement, prepend_scheme=prepend_scheme))
         return pre_tokenizers.Sequence(list_pretokenizers)
 
     def normalizer(self, proto):
@@ -1007,10 +1009,11 @@ class PegasusConverter(SpmConverter):
         return proto.trainer_spec.unk_id + self.original_tokenizer.offset
 
     def pre_tokenizer(self, replacement, add_prefix_space):
+        prepend_scheme = "always" if add_prefix_space else "never"
         return pre_tokenizers.Sequence(
             [
                 pre_tokenizers.WhitespaceSplit(),
-                pre_tokenizers.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space),
+                pre_tokenizers.Metaspace(replacement=replacement, prepend_scheme=prepend_scheme),
             ]
         )
 
diff --git a/src/transformers/dependency_versions_table.py b/src/transformers/dependency_versions_table.py
index d70b717f0d..42c09fdef2 100644
--- a/src/transformers/dependency_versions_table.py
+++ b/src/transformers/dependency_versions_table.py
@@ -79,7 +79,7 @@ deps = {
     "tf2onnx": "tf2onnx",
     "timeout-decorator": "timeout-decorator",
     "timm": "timm",
-    "tokenizers": "tokenizers>=0.14,<0.19",
+    "tokenizers": "tokenizers>=0.16,<0.19",
     "torch": "torch",
     "torchaudio": "torchaudio",
     "torchvision": "torchvision",