llama-r/special_tokens_map.json

{
  "bos_token": "<|sep|>",
  "eos_token": "<|endoftext|>",
  "pad_token": "<|endoftext|>"
}