Mitssuki-Sakamoto's picture
add highdataset config to README
995aa68
metadata
license: cc-by-nc-4.0
dataset_info:
  - config_name: 1.4b_mix_seed1
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
      - name: index
        dtype: int64
    splits:
      - name: preference
        num_bytes: 25889425.028748564
        num_examples: 20000
    download_size: 12367081
    dataset_size: 25889425.028748564
  - config_name: 160m_mix_seed1
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
      - name: index
        dtype: int64
    splits:
      - name: preference
        num_bytes: 25900235.98820059
        num_examples: 20000
    download_size: 12316110
    dataset_size: 25900235.98820059
  - config_name: 1.4b_mix_seed2
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
      - name: index
        dtype: int64
    splits:
      - name: preference
        num_bytes: 25889425.028748564
        num_examples: 20000
    download_size: 12359463
    dataset_size: 25889425.028748564
  - config_name: 160m_mix_seed2
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
      - name: index
        dtype: int64
    splits:
      - name: preference
        num_bytes: 25900235.98820059
        num_examples: 20000
    download_size: 12313149
    dataset_size: 25900235.98820059
  - config_name: 1.4b_mix_seed3
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
      - name: index
        dtype: int64
    splits:
      - name: preference
        num_bytes: 25889425.028748564
        num_examples: 20000
    download_size: 12358176
    dataset_size: 25889425.028748564
  - config_name: 160m_mix_seed3
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
      - name: index
        dtype: int64
    splits:
      - name: preference
        num_bytes: 25900235.98820059
        num_examples: 20000
    download_size: 12311452
    dataset_size: 25900235.98820059
  - config_name: 1.4b_high
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
    splits:
      - name: preference
        num_bytes: 25791877
        num_examples: 20001
    download_size: 12310829
    dataset_size: 25791877
  - config_name: 160m_high
    features:
      - name: instruction
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
      - name: preference
        dtype: int64
      - name: output_1
        dtype: string
      - name: output_2
        dtype: string
      - name: reward_model_prompt_format
        dtype: string
      - name: gen_prompt_format
        dtype: string
      - name: gen_kwargs
        struct:
          - name: do_sample
            dtype: bool
          - name: max_new_tokens
            dtype: int64
          - name: pad_token_id
            dtype: int64
          - name: top_k
            dtype: int64
          - name: top_p
            dtype: float64
      - name: reward_1
        dtype: float64
      - name: reward_2
        dtype: float64
      - name: n_samples
        dtype: int64
      - name: reject_select
        dtype: string
    splits:
      - name: preference
        num_bytes: 25837484
        num_examples: 20001
    download_size: 12262392
    dataset_size: 25837484
configs:
  - config_name: 1.4b_mix_seed1
    data_files:
      - split: preference
        path: 1.4b_mix_seed1/preference-*
  - config_name: 160m_mix_seed1
    data_files:
      - split: preference
        path: 160m_mix_seed1/preference-*
  - config_name: 1.4b_mix_seed2
    data_files:
      - split: preference
        path: 1.4b_mix_seed2/preference-*
  - config_name: 160m_mix_seed2
    data_files:
      - split: preference
        path: 160m_mix_seed2/preference-*
  - config_name: 1.4b_mix_seed3
    data_files:
      - split: preference
        path: 1.4b_mix_seed3/preference-*
  - config_name: 160m_mix_seed3
    data_files:
      - split: preference
        path: 160m_mix_seed3/preference-*
  - config_name: 1.4b_high
    data_files:
      - split: preference
        path: 1.4b_high/preference-*
  - config_name: 160m_high
    data_files:
      - split: preference
        path: 160m_high/preference-*

This work utilizes the alpaca_farm dataset provided by Tatsu-lab, which is available at https://huggingface.co/datasets/tatsu-lab/alpaca_farm and is used under the CC BY-NC 4.0 license.