fix: LlaMATokenizer -> LlamaTokenizer

It's very confusing, but in transformers >= 4.28.0, it is spelled as "Llama" rather than "LlaMA" as in:
https://github.com/huggingface/transformers/blob/32ff06403d48fa018a33ccab02a5f9d3eb8b078b/src/transformers/__init__.py#L4320

tokenizer_config.json

@@ -1 +1 @@
-{"bos_token": "", "eos_token": "", "model_max_length": 1000000000000000019884624838656, "tokenizer_class": "LlaMATokenizer", "unk_token": ""}
+{"bos_token": "", "eos_token": "", "model_max_length": 1000000000000000019884624838656, "tokenizer_class": "LlamaTokenizer", "unk_token": ""}