rbiswasfc/reward-bench-binary-classification

---
license: odc-by
dataset_info:
features:
- name: id
dtype: int64
- name: prompt
dtype: string
- name: response_a
dtype: string
- name: response_b
dtype: string
- name: label
dtype: int64
- name: subset
dtype: string
splits:
- name: train
num_bytes: 7856927
num_examples: 3845
- name: test
num_bytes: 2877668
num_examples: 1278
download_size: 5663108
dataset_size: 10734595
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
---

Repurposed `allenai/reward-bench` dataset for binary classification task, using the following script:
```python
import random

import pandas as pd
from datasets import Dataset, load_dataset
from sklearn.model_selection import GroupKFold

data_df = load_dataset("allenai/reward-bench", split="raw").to_pandas()

rng = random.Random(43)

examples = []
for idx, row in data_df.iterrows():
if rng.random() > 0.5:
response_a = row["chosen"]
response_b = row["rejected"]
label = 0
else:
response_a = row["rejected"]
response_b = row["chosen"]
label = 1

this_ex = {
"id": row["id"],
"prompt": row["prompt"],
"response_a": response_a,
"response_b": response_b,
"label": label,
"subset": row["subset"],
}

examples.append(this_ex)
processed_df = pd.DataFrame(examples)

# train-test split

def add_fold(df, n_splits=4, group_col="subset"):
skf = GroupKFold(n_splits=n_splits)

for f, (t_, v_) in enumerate(skf.split(df, y=df[group_col].values, groups=df[group_col].values)):
df.loc[v_, "kfold"] = f
return df

processed_df = add_fold(processed_df)

test_fold = 0

test_df = processed_df[processed_df["kfold"] == test_fold].drop(columns="kfold")
train_df = processed_df[processed_df["kfold"] != test_fold].drop(columns="kfold")

train_ds = Dataset.from_pandas(train_df, preserve_index=False)
test_ds = Dataset.from_pandas(test_df, preserve_index=False)
```

原始信息汇总

数据集概述

数据集特征

• id: 整数类型 (int64)
• prompt: 字符串类型 (string)
• response_a: 字符串类型 (string)
• response_b: 字符串类型 (string)
• label: 整数类型 (int64)
• subset: 字符串类型 (string)

数据集分割

• 训练集:
  • 示例数量: 3845
  • 数据大小: 7856927 字节
• 测试集:
  • 示例数量: 1278
  • 数据大小: 2877668 字节

数据集大小

• 下载大小: 5663108 字节
• 数据集总大小: 10734595 字节

配置文件

• 默认配置:
  • 训练数据路径: data/train-*
  • 测试数据路径: data/test-*