Closed ChengMingZhang-ZTE closed 8 months ago
I have same question. When pre-training llama2(code llama), which one is used as the delimiter between samples: eos or bos?
For training, we add both BOS and EOS tokens.
@jgehring I understand, thanks for your reply.
I am curious about the form of the dataset for Code Llama pre-training. I want to know whether eos or bos was used during the pre-training process. For example, the data format is {code}{EOS} or {BOS}{code}, which format is used for Code Llama pre-training?