RLHFlow/HH-RLHF-Helpful-standard

---
dataset_info:
features:
- name: rejected_score
dtype: 'null'
- name: chosen
list:
- name: content
dtype: string
- name: role
dtype: string
- name: rejected
list:
- name: content
dtype: string
- name: role
dtype: string
- 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-*
---

We process the helpful subset of Anthropic-HH into the standard format. The filtering script is as follows.

```python

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
```

原始信息汇总

数据集概述

数据集特征

• rejected_score：数据类型为 null。
• chosen：包含两个子特征
  • content：数据类型为 string。
  • role：数据类型为 string。
• rejected：包含两个子特征
  • content：数据类型为 string。
  • role：数据类型为 string。
• chosen_score：数据类型为 null。

数据集划分

• train：包含115396个示例，数据集大小为248008356.9440396字节。

数据集大小

• 下载大小：139765640字节。
• 数据集总大小：248008356.9440396字节。

配置

• config_name: default
  • data_files:
    • split: train
      • path: data/train-*