metadata
dataset_info:
features:
- name: idx
dtype: float64
- name: label
dtype: large_string
- name: source
dtype: large_string
- name: prompt
dtype: large_string
- name: response
dtype: large_string
- name: overall_influence_direction
dtype: large_string
- name: overall_influence_sum
dtype: float64
- name: more_negative_than_overall_median_for_harmful
dtype: bool
- name: more_negative_than_harmful_median_for_harmful
dtype: bool
- name: more_negative_than_harmless_median_for_harmful
dtype: bool
- name: more_positive_than_overall_median_for_harmless
dtype: bool
- name: more_positive_than_harmful_median_for_harmless
dtype: bool
- name: more_positive_than_harmless_median_for_harmless
dtype: bool
- name: top_5_positive
dtype: large_string
- name: top_5_negative
dtype: large_string
- name: Darian_Misalligned_qual
dtype: large_string
- name: Darian Why
dtype: large_string
- name: misalligned_quant
dtype: bool
splits:
- name: train
num_bytes: 231016
num_examples: 113
download_size: 84522
dataset_size: 231016
configs:
- config_name: default
data_files:
- split: train
path: data/train-*