metadata
dataset_info:
features:
- name: prompt_id
dtype: string
- name: prompt
dtype: string
- name: all_generated_responses
sequence: string
- name: all_rm_scores
sequence: float64
- name: chosen
list:
- name: content
dtype: string
- name: role
dtype: string
- name: rejected
list:
- name: content
dtype: string
- name: role
dtype: string
- name: ppl_chosen
dtype: float64
- name: nll_chosen
dtype: float64
- name: ppl_rejected
dtype: float64
- name: nll_rejected
dtype: float64
splits:
- name: train
num_bytes: 792097911
num_examples: 59569
- name: test
num_bytes: 26136876
num_examples: 1941
- name: train_filtered_full
num_bytes: 780329933.5077641
num_examples: 58684
- name: train_largest_30_ppl_rejected
num_bytes: 234096320.62238747
num_examples: 17605
- name: train_largest_30_ppl_chosen
num_bytes: 234096320.62238747
num_examples: 17605
- name: train_largest_30_abs_diff
num_bytes: 236740711.52869606
num_examples: 17605
- name: train_smallest_30_abs_diff
num_bytes: 236740711.52869606
num_examples: 17605
- name: train_large_correct
num_bytes: 236740711.52869606
num_examples: 17605
- name: train_large_wrong
num_bytes: 236740711.52869606
num_examples: 17605
- name: train_small_correct
num_bytes: 236740711.52869606
num_examples: 17605
- name: train_small_wrong
num_bytes: 236740711.52869606
num_examples: 17605
download_size: 1644165081
dataset_size: 3487201630.924715
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
- split: train_filtered_full
path: data/train_filtered_full-*
- split: train_largest_30_ppl_rejected
path: data/train_largest_30_ppl_rejected-*
- split: train_largest_30_ppl_chosen
path: data/train_largest_30_ppl_chosen-*
- split: train_largest_30_abs_diff
path: data/train_largest_30_abs_diff-*
- split: train_smallest_30_abs_diff
path: data/train_smallest_30_abs_diff-*
- split: train_large_correct
path: data/train_large_correct-*
- split: train_large_wrong
path: data/train_large_wrong-*
- split: train_small_correct
path: data/train_small_correct-*
- split: train_small_wrong
path: data/train_small_wrong-*