val-bench/calibration-unaligned-0

dataset_info:
features:

• name: article_title
  dtype: string
• name: url
  dtype: string
• name: section_title
  dtype: string
• name: section_text
  dtype: string
• name: issue_scale
  dtype: int64
• name: for_question
  dtype: string
• name: against_question
  dtype: string
• name: id
  dtype: int64
• name: category
  dtype: string
• name: issue_description
  dtype: string
• name: for_prompt
  dtype: string
• name: against_prompt
  dtype: string
• name: for_response
  struct:
  • name: content
    dtype: string
  • name: role
    dtype: string
• name: stop_reason
  dtype: string
• name: finish_reason
  dtype: string
• name: against_response
  struct:
  • name: content
    dtype: string
  • name: role
    dtype: string
• name: evaluation_prompt
  list:
  • name: content
    dtype: string
  • name: role
    dtype: string
• name: alignment_judge_response
  struct:
  • name: content
    dtype: string
  • name: role
    dtype: string
• name: alignment_score
  dtype: int64
• name: alignment_judge_parsed
  struct:
  • name: alignment_score
    dtype: int64
  • name: key_points_of_agreement
    dtype: string
  • name: key_points_of_disagreement
    dtype: string
  • name: person_1_core_position
    dtype: string
  • name: person_1_has_no_information
    dtype: bool
  • name: person_1_refusal
    dtype: bool
  • name: person_2_core_position
    dtype: string
  • name: person_2_has_no_information
    dtype: bool
  • name: person_2_refusal
    dtype: bool
    splits:
• name: test
  num_bytes: 186728410
  num_examples: 11475
  download_size: 77190126
  dataset_size: 186728410
  configs:
• config_name: default
  data_files:
  • split: test
    path: data/test-*