tokenizer_config.json

{
    "backend": "tokenizers",
    "bos_token": "<|startoftext|>",
    "clean_up_tokenization_spaces": false,
    "eos_token": "<|return|>",
    "is_local": false,
    "model_input_names": [
        "input_ids",
        "attention_mask"
    ],
    "model_max_length": 131072,
    "pad_token": "<|endoftext|>",
    "tokenizer_class": "PreTrainedTokenizerFast"
}