118 lines
3.2 KiB
Markdown
118 lines
3.2 KiB
Markdown
# HellaSwag
|
||
|
||
|
||
## 概述
|
||
|
||
HellaSwag 是一个用于评估常识性自然语言推理的基准测试，专门测试模型完成描述日常情境句子的能力。该数据集采用对抗性过滤（adversarial filtering）方法生成具有挑战性的干扰项，这些干扰项在语法上正确但在语义上不合理。
|
||
|
||
## 任务描述
|
||
|
||
- **任务类型**：多项选择题句子补全
|
||
- **输入**：描述某项活动或情境的上下文
|
||
- **输出**：从 4 个选项（A、B、C、D）中选择最合理的后续内容
|
||
- **领域**：日常活动与常识场景
|
||
|
||
## 主要特点
|
||
|
||
- 包含 70,000 多个问题，用于测试基于现实情境的常识推理能力
|
||
- 上下文来源于 ActivityNet 和 WikiHow
|
||
- 错误选项经过对抗性过滤处理
|
||
- 要求理解典型事件序列
|
||
- 测试物理常识与社会常识推理能力
|
||
|
||
## 评估说明
|
||
|
||
- 默认配置使用 **0-shot** 评估方式
|
||
- 在验证集（validation split）上进行评估
|
||
- 已对选项结尾进行预处理，清除格式化伪影
|
||
- 上下文由 `ctx_a` 和 `ctx_b` 字段拼接而成
|
||
- 元数据中包含活动标签，可用于分析
|
||
|
||
## 属性
|
||
|
||
| 属性 | 值 |
|
||
|----------|-------|
|
||
| **基准测试名称** | `hellaswag` |
|
||
| **数据集ID** | [evalscope/hellaswag](https://modelscope.cn/datasets/evalscope/hellaswag/summary) |
|
||
| **论文** | N/A |
|
||
| **标签** | `Commonsense`, `Knowledge`, `MCQ` |
|
||
| **指标** | `acc` |
|
||
| **默认示例数量** | 0-shot |
|
||
| **评估划分** | `validation` |
|
||
|
||
|
||
## 数据统计
|
||
|
||
| 指标 | 值 |
|
||
|--------|-------|
|
||
| 总样本数 | 10,042 |
|
||
| 提示词长度（平均） | 767.6 字符 |
|
||
| 提示词长度（最小/最大） | 329 / 1655 字符 |
|
||
|
||
## 样例示例
|
||
|
||
**子集**: `default`
|
||
|
||
```json
|
||
{
|
||
"input": [
|
||
{
|
||
"id": "18df47e2",
|
||
"content": "Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of A,B,C,D.\n\nA man is sitting on a roof. He\n\nA) is using wrap to wrap a pair of skis.\nB) is ripping level tiles off.\nC) is holding a rubik's cube.\nD) starts pulling up roofing on a roof."
|
||
}
|
||
],
|
||
"choices": [
|
||
"is using wrap to wrap a pair of skis.",
|
||
"is ripping level tiles off.",
|
||
"is holding a rubik's cube.",
|
||
"starts pulling up roofing on a roof."
|
||
],
|
||
"target": "D",
|
||
"id": 0,
|
||
"group_id": 0,
|
||
"metadata": {
|
||
"activity_label": "Roof shingle removal"
|
||
}
|
||
}
|
||
```
|
||
|
||
## 提示模板
|
||
|
||
**提示模板：**
|
||
```text
|
||
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
|
||
|
||
{question}
|
||
|
||
{choices}
|
||
```
|
||
|
||
## 使用方法
|
||
|
||
### 使用命令行（CLI）
|
||
|
||
```bash
|
||
evalscope eval \
|
||
--model YOUR_MODEL \
|
||
--api-url OPENAI_API_COMPAT_URL \
|
||
--api-key EMPTY_TOKEN \
|
||
--datasets hellaswag \
|
||
--limit 10 # 正式评估时请删除此行
|
||
```
|
||
|
||
### 使用 Python
|
||
|
||
```python
|
||
from evalscope import run_task
|
||
from evalscope.config import TaskConfig
|
||
|
||
task_cfg = TaskConfig(
|
||
model='YOUR_MODEL',
|
||
api_url='OPENAI_API_COMPAT_URL',
|
||
api_key='EMPTY_TOKEN',
|
||
datasets=['hellaswag'],
|
||
limit=10, # 正式评估时请删除此行
|
||
)
|
||
|
||
run_task(task_cfg=task_cfg)
|
||
``` |