dataset_info:
features:
- name: rejected_score
dtype: 'null' - name: chosen
list:- name: content
dtype: string - name: role
dtype: string
- name: content
- name: rejected
list:- name: content
dtype: string - name: role
dtype: string
- name: content
- name: chosen_score
dtype: 'null'
splits: - name: train
num_bytes: 248008356.9440396
num_examples: 115396
download_size: 139765640
dataset_size: 248008356.9440396
configs: - config_name: default
data_files:- split: train
path: data/train-*
- split: train
We process the helpful subset of Anthropic-HH into the standard format. The filtering script is as follows.
def filter_example(example):
if len(example['chosen']) != len(example['rejected']):
return False
if len(example['chosen']) % 2 != 0:
return False
n_rounds = len(example['chosen'])
for i in range(len(example['chosen'])):
if example['chosen'][i]['role'] != ['user', 'assistant'][i % 2]:
return False
if example['rejected'][i]['role'] != ['user', 'assistant'][i % 2]:
return False
if len(example['chosen'][i]['content']) == 0:
return False
if len(example['rejected'][i]['content']) == 0:
return False
if i < n_rounds - 1:
if example['chosen'][i]['content'] != example['rejected'][i]['content']:
return False
else:
if example['chosen'][i]['content'] == example['rejected'][i]['content']:
return False
return True