Support for Llama 3.1, 3.2, and 3.3

[urroxyz]

I'm creating an issue for this so we can keep track of it.

PreTrainedTokenizerFast is currently unsupported.

@Saibo-creator, I'm not quite sure how new tokenizer support is added. Could you give me the rundown?

[Saibo-creator]

PreTrainedTokenizerFast should already be supported:
https://github.com/epfl-dlab/transformers-CFG/blob/main/transformers_cfg/tokenization/SUPPORTED_TOKENIZERS.py#L7

I think Llama 3.1, 3.2, and 3.3 use the same tokenizer as Llama 3, so supporting them should be pretty straightforward.

I’m taking a look now.