metadata
dataset_info:
features:
- name: text
dtype: string
- name: meta
struct:
- name: pile_set_name
dtype: string
- name: sentences_perturbed
dtype: int64
- name: length_stats
struct:
- name: length_ratios
list: float64
- name: original_lengths
list: int64
- name: paraphrase_lengths
list: int64
- name: log_data
struct:
- name: alpha
dtype: float64
- name: baseline_model
dtype: string
- name: document_id
dtype: int64
- name: most_advantageous_only
dtype: int64
- name: original_text
dtype: string
- name: perturbed_text
dtype: string
- name: sentence_changes
list:
- name: available_paraphrases
list: string
- name: final_paraphrase
dtype: string
- name: left_context
dtype: string
- name: original_sentence
dtype: string
- name: right_context
dtype: string
- name: selected_paraphrase
dtype: string
- name: sentence_index
dtype: int64
- name: target
dtype: string
- name: token_losses
dtype: 'null'
- name: was_perturbed
dtype: bool
- name: sentences_perturbed
dtype: int64
- name: target_model
dtype: string
- name: threshold
dtype: float64
splits:
- name: validation
num_bytes: 24824365
num_examples: 2000
download_size: 13304805
dataset_size: 24824365
configs:
- config_name: default
data_files:
- split: validation
path: data/validation-*