tokenizer_config.json

{
  "tokenizer_class": "PreTrainedTokenizerFast",
  "bos_token": "[BOS]",
  "eos_token": "<|endoftext|>",
  "unk_token": "[UNK]",
  "pad_token": "[PAD]",
  "additional_special_tokens": [
    "<|user|>",
    "<|assistant|>",
    "<|fim_prefix|>",
    "<|fim_middle|>",
    "<|fim_suffix|>"
  ],
  "model_max_length": 32768,
  "clean_up_tokenization_spaces": false
}