数据集:

lucasmccabe-lmi/FLAN_CoT_alpaca_style

预印本库:

arxiv:2210.11416
中文

Dataset Card for "FLAN_CoT_alpaca_style"

We provide a dataset representing the 9 chain-of-thought (reasoning) fine-tuning tasks from FLAN . Minor formatting has been applied:

  • We apply an Alpaca-style format (i.e. instruction/input/output fields)
  • If the question is multiple-choice, the options are provided in the input field
  • The phrase "Explain your reasoning step-by-step before providing the correct answer." is added to the end of the instruction field.

Numbers:

Prompts: 74771

Tokens: 9016176 using the EleutherAI/gpt-neox-20b tokenizer (counting instruction+input+output)