112 lines
2.8 KiB
Markdown
112 lines
2.8 KiB
Markdown
# SIQA
|
||
|
||
## 概述
|
||
|
||
SIQA（Social Interaction QA）是一个用于评估社会常识智能的基准测试，旨在衡量模型对人类行为及其社会影响的理解能力。与侧重于物理知识的基准不同，SIQA 专注于对人类行为的推理。
|
||
|
||
## 任务描述
|
||
|
||
- **任务类型**：社会常识推理
|
||
- **输入**：包含社交情境的上下文、一个问题以及三个选项
|
||
- **输出**：最符合社会规范的答案（A、B 或 C）
|
||
- **重点**：人类行为、动机及其社会影响
|
||
|
||
## 主要特点
|
||
|
||
- 测试社会智能与情感理解能力
|
||
- 问题围绕人类行为及其后果展开
|
||
- 涵盖动机、反应和社会规范
|
||
- 包含超过 33,000 个众包问答对
|
||
- 需要对人类心理进行推理
|
||
|
||
## 评估说明
|
||
|
||
- 默认配置使用 **0-shot** 评估
|
||
- 使用简单的多项选择提示方式
|
||
- 在验证集（validation split）上进行评估
|
||
- 使用简单准确率（accuracy）作为评估指标
|
||
|
||
## 属性
|
||
|
||
| 属性 | 值 |
|
||
|----------|-------|
|
||
| **基准测试名称** | `siqa` |
|
||
| **数据集 ID** | [extraordinarylab/siqa](https://modelscope.cn/datasets/extraordinarylab/siqa/summary) |
|
||
| **论文** | N/A |
|
||
| **标签** | `Commonsense`, `MCQ`, `Reasoning` |
|
||
| **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 | 1,954 |
|
||
| 提示词长度（平均） | 289.05 字符 |
|
||
| 提示词长度（最小/最大） | 242 / 509 字符 |
|
||
|
||
## 样例示例
|
||
|
||
**子集**: `default`
|
||
|
||
```json
|
||
{
|
||
"input": [
|
||
{
|
||
"id": "8d09aab2",
|
||
"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.\n\nWhat does Tracy need to do before this?\n\nA) make a new plan\nB) Go home and see Riley\nC) Find somewhere to go"
|
||
}
|
||
],
|
||
"choices": [
|
||
"make a new plan",
|
||
"Go home and see Riley",
|
||
"Find somewhere to go"
|
||
],
|
||
"target": "C",
|
||
"id": 0,
|
||
"group_id": 0,
|
||
"metadata": {}
|
||
}
|
||
```
|
||
|
||
## 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 CLI
|
||
|
||
```bash
|
||
evalscope eval \
|
||
--model YOUR_MODEL \
|
||
--api-url OPENAI_API_COMPAT_URL \
|
||
--api-key EMPTY_TOKEN \
|
||
--datasets siqa \
|
||
--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=['siqa'],
|
||
limit=10, # 正式评估时请删除此行
|
||
)
|
||
|
||
run_task(task_cfg=task_cfg)
|
||
``` |