Save the vllm tokenizer adapted state

Author: mory91

outlines/serve/vllm.py

@@ -48,6 +48,9 @@ def _adapt_tokenizer(tokenizer):
     Llama's tokenizer to be able to compile FSMs for this model.
 
     """
+    if getattr(tokenizer, "_outlines_adapted", False):
+        return tokenizer
+
     tokenizer.vocabulary = tokenizer.get_vocab()
     tokenizer.special_tokens = set(tokenizer.all_special_tokens)
 
@@ -72,6 +75,7 @@ def new_decoder(inp_tokens: List[int]) -> List[str]:
 
     tokenizer.convert_token_to_string = convert_token_to_string
     tokenizer.decode = change_decoder(tokenizer.decode)
+    setattr(tokenizer, "_outlines_adapted", True)
 
     return tokenizer