README.md

metadata

dataset_info:
  - config_name: R_SFT
    features:
      - name: id
        dtype: string
      - name: conversations
        list:
          - name: role
            dtype: string
          - name: content
            dtype: string
      - name: prompt
        dtype: string
      - name: answer
        dtype: string
      - name: source
        dtype: string
    splits:
      - name: train
        num_bytes: 9202798
        num_examples: 1000
    download_size: 4619497
    dataset_size: 9202798
  - config_name: SFT
    features:
      - name: id
        dtype: string
      - name: conversations
        list:
          - name: role
            dtype: string
          - name: content
            dtype: string
      - name: source
        dtype: string
    splits:
      - name: train
        num_bytes: 3465754
        num_examples: 1030
    download_size: 1831048
    dataset_size: 3465754
  - config_name: U_SFT
    features:
      - name: id
        dtype: string
      - name: conversations
        list:
          - name: role
            dtype: string
          - name: content
            dtype: string
      - name: prompt
        dtype: string
      - name: answer
        dtype: string
      - name: source
        dtype: string
    splits:
      - name: train
        num_bytes: 7573548
        num_examples: 1000
    download_size: 3894316
    dataset_size: 7573548
configs:
  - config_name: R_SFT
    data_files:
      - split: train
        path: R_SFT/train-*
  - config_name: SFT
    data_files:
      - split: train
        path: SFT/train-*
  - config_name: U_SFT
    data_files:
      - split: train
        path: U_SFT/train-*

SFT: default LIMA
Update_SFT: gpt-4o-mini-2024-07-18으로 default LIMA의 prompt만 활용해 answer를 출력
R_SFT: 대충 아이디어 실험할려고 만든 데이터, U_SFT와 동일하게 gpt-4o-mini-2024-07-18로 생성
멀티턴엔 따로 증강 안했음.