[Bugfix] Cache added_vocab to avoid per-token overhead (#30351)

Signed-off-by: limingliang <limingliang@stepfun.com>
Co-authored-by: limingliang <limingliang@stepfun.com>

vllm/tokenizers/deepseekv32.py

@@ -17,6 +17,8 @@ class DeepseekV32Tokenizer(HfTokenizer):
         self.name_or_path = (
             tokenizer.name_or_path if hasattr(tokenizer, "name_or_path") else ""
         )
+        self._added_vocab = self.tokenizer.get_added_vocab()
+        self._added_vocab_size = len(self._added_vocab)
 
     @classmethod
     def from_pretrained(
@@ -98,7 +100,7 @@ class DeepseekV32Tokenizer(HfTokenizer):
 
     def __len__(self) -> int:
         # </think> is an added token in DeepseekV32 tokenizer
-        return self.vocab_size + len(self.get_added_vocab())
+        return self.vocab_size + self._added_vocab_size
 
     def __call__(
         self,
@@ -120,7 +122,7 @@ class DeepseekV32Tokenizer(HfTokenizer):
         return self.tokenizer.get_vocab()
 
     def get_added_vocab(self) -> dict[str, int]:
-        return self.tokenizer.get_added_vocab()
+        return self._added_vocab.copy()
 
     def encode(
         self,