tokenizer_config.json

{
  "add_prefix_space": false,
  "bos_token": "<|endoftext|>",
  "clean_up_tokenization_spaces": true,
  "eos_token": "<|endoftext|>",
  "model_max_length": 16384,
  "tokenizer_class": "GPT2TokenizerFast",
  "unk_token": "<|endoftext|>",
  "chat_template": "{% for message in messages %}{{'<|im_start|>'+ message['role'] + \n'+ message['content'] + '\n<|im_end|>\n'}}{% endfor %}{{'<|im_start|>assistant\n'}}"
}