Update tokenizer_pi05.py

tokenizer_pi05.py

@@ -14,14 +14,14 @@ class PaligemmaTokenizer(PreTrainedTokenizer):
     """
 
     vocab_files_names = {"vocab_file": "tokenizer.model"}
-    model_input_names = ["input_ids", "attention_mask"]
 
     def __init__(self, vocab_file: str, max_len: int = 48, **kwargs):
-        super().__init__(**kwargs)
-
+        # Must init _tokenizer BEFORE super().__init__ because parent accesses vocab_size
         self.vocab_file = vocab_file
         self._max_len = int(max_len)
         self._tokenizer = sentencepiece.SentencePieceProcessor(model_file=str(vocab_file))
+
+        super().__init__(**kwargs)
         self.pad_token_id = 0
 
     # ---- minimal HF plumbing ----