AlignmentResearch/XSTest

dataset_info:

• config_name: default
  features:
  • name: clf_label
    dtype:
    class_label:
    names:
    '0': Benign
    '1': Harmful
  • name: instructions
    dtype: string
  • name: content
    sequence: string
  • name: answer_prompt
    dtype: string
  • name: proxy_clf_label
    dtype: int64
  • name: gen_target
    dtype: string
  • name: proxy_gen_target
    dtype: string
    splits:
  • name: validation
    num_bytes: 43753
    num_examples: 450
  • name: train
    num_bytes: 0
    num_examples: 0
    download_size: 18281
    dataset_size: 43753
• config_name: neg
  features:
  • name: clf_label
    dtype:
    class_label:
    names:
    '0': Benign
    '1': Harmful
  • name: instructions
    dtype: string
  • name: content
    sequence: string
  • name: answer_prompt
    dtype: string
  • name: proxy_clf_label
    dtype: int64
  • name: gen_target
    dtype: string
  • name: proxy_gen_target
    dtype: string
    splits:
  • name: validation
    num_bytes: 24307.222222222223
    num_examples: 250
  • name: train
    num_bytes: 0
    num_examples: 0
    download_size: 14192
    dataset_size: 24307.222222222223
• config_name: pos
  features:
  • name: clf_label
    dtype:
    class_label:
    names:
    '0': Benign
    '1': Harmful
  • name: instructions
    dtype: string
  • name: content
    sequence: string
  • name: answer_prompt
    dtype: string
  • name: proxy_clf_label
    dtype: int64
  • name: gen_target
    dtype: string
  • name: proxy_gen_target
    dtype: string
    splits:
  • name: validation
    num_bytes: 19445.777777777777
    num_examples: 200
  • name: train
    num_bytes: 0
    num_examples: 0
    download_size: 12209
    dataset_size: 19445.777777777777
    configs:
• config_name: default
  data_files:
  • split: validation
    path: data/validation-*
  • split: train
    path: data/train-*
• config_name: neg
  data_files:
  • split: validation
    path: neg/validation-*
  • split: train
    path: neg/train-*
• config_name: pos
  data_files:
  • split: validation
    path: pos/validation-*
  • split: train
    path: pos/train-*