[Lllama] Update tokenization code to ensure parsing of the special tokens [core] (#24042)

* preventllama fast from returning token type ids

* remove type hints

* normalised False

src/transformers/convert_slow_tokenizer.py

@@ -1134,9 +1134,9 @@ class LlamaConverter(SpmConverter):
             )
             tokenizer.add_special_tokens(
                 [
-                    AddedToken("<unk>", normalized=True),
-                    AddedToken("<s>", normalized=True),
-                    AddedToken("</s>", normalized=True),
+                    AddedToken("<unk>", normalized=False),
+                    AddedToken("<s>", normalized=False),
+                    AddedToken("</s>", normalized=False),
                 ]
             )
         else:

src/transformers/models/llama/tokenization_llama_fast.py

@@ -77,6 +77,7 @@ class LlamaTokenizerFast(PreTrainedTokenizerFast):
     vocab_files_names = VOCAB_FILES_NAMES
     slow_tokenizer_class = LlamaTokenizer
     padding_side = "left"
+    model_input_names = ["input_ids", "attention_mask"]
 
     def __init__(
         self,