metadata
dataset_info:
features:
- name: instruction
dtype: string
- name: response_accepted
dtype: string
- name: response_rejected
dtype: string
- name: instruction_perplexity_kenlm
dtype: int64
- name: chosen_response_perplexity_kenlm
dtype: int64
- name: rejected_response_perplexity_kenlm
dtype: int64
- name: combined_perplexity_dpo
dtype: int64
- name: combined_perplexity_sft
dtype: int64
- name: instruction_lang
dtype: string
- name: instruction_lang_proba
dtype: float64
- name: chosen_response_lang
dtype: string
- name: chosen_response_lang_proba
dtype: float64
- name: rejected_response_lang
dtype: string
- name: rejected_response_lang_proba
dtype: float64
- name: perplexity_instruction_len_ratio
dtype: float64
- name: perplexity_response_len_ratio
dtype: float64
- name: dataset_source
dtype: string
- name: __index_level_0__
dtype: int64
splits:
- name: train
num_bytes: 11944407
num_examples: 4402
download_size: 7055078
dataset_size: 11944407
configs:
- config_name: default
data_files:
- split: train
path: data/train-*