metadata
dataset_info:
features:
- name: text
dtype: string
- name: meta
struct:
- name: pile_set_name
dtype: string
- name: change_metrics
struct:
- name: baseline_loss
dtype: float32
- name: char_change_proportion
dtype: float64
- name: length_change_ratio
dtype: float64
- name: length_difference
dtype: int64
- name: sentence_change_proportion
dtype: float64
- name: sentences_changed_count
dtype: int64
- name: sentences_changed_ratio
dtype: float64
- name: sentences_considered
dtype: int64
- name: sentences_perturbed
dtype: int64
- name: target_loss
dtype: float32
- name: token_change_proportion
dtype: float64
- name: total_original_length
dtype: int64
- name: total_perturbed_length
dtype: int64
- name: advantage_metrics
struct:
- name: avg_advantage_score
dtype: float64
- name: max_advantage_score
dtype: float64
- name: sentences_with_advantage
dtype: int64
- name: total_advantage_score
dtype: float64
splits:
- name: validation
num_bytes: 15449505
num_examples: 2000
download_size: 8155635
dataset_size: 15449505
configs:
- config_name: default
data_files:
- split: validation
path: data/validation-*