113 lines
2.7 KiB
Markdown
113 lines
2.7 KiB
Markdown
# SciQ
|
||
|
||
## 概述
|
||
|
||
SciQ 是一个众包的科学考试题目数据集，涵盖物理、化学、生物及其他科学领域。大多数题目包含支持性的证据段落。
|
||
|
||
## 任务描述
|
||
|
||
- **任务类型**：科学问答（多项选择题）
|
||
- **输入**：一道包含 4 个选项的科学问题
|
||
- **输出**：正确答案的字母（A、B、C 或 D）
|
||
- **领域**：物理、化学、生物、地球科学等
|
||
|
||
## 主要特点
|
||
|
||
- 众包收集的科学考试题目
|
||
- 覆盖多个科学领域
|
||
- 提供支持性证据段落
|
||
- 考察科学知识与推理能力
|
||
- 适用于科学理解能力评估
|
||
|
||
## 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
|
||
- 在测试集（test 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** | `sciq` |
|
||
| **数据集ID** | [extraordinarylab/sciq](https://modelscope.cn/datasets/extraordinarylab/sciq/summary) |
|
||
| **论文** | N/A |
|
||
| **标签** | `Knowledge`, `MCQ`, `ReadingComprehension` |
|
||
| **指标** | `acc` |
|
||
| **默认示例数量** | 0-shot |
|
||
| **评估划分** | `test` |
|
||
|
||
## 数据统计
|
||
|
||
| 指标 | 值 |
|
||
|--------|-------|
|
||
| 总样本数 | 1,000 |
|
||
| 提示词长度（平均） | 322.68 字符 |
|
||
| 提示词长度（最小/最大） | 244 / 505 字符 |
|
||
|
||
## 样例示例
|
||
|
||
**子集**: `default`
|
||
|
||
```json
|
||
{
|
||
"input": [
|
||
{
|
||
"id": "a30097af",
|
||
"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\nCompounds that are capable of accepting electrons, such as o 2 or f2, are called what?\n\nA) antioxidants\nB) Oxygen\nC) residues\nD) oxidants"
|
||
}
|
||
],
|
||
"choices": [
|
||
"antioxidants",
|
||
"Oxygen",
|
||
"residues",
|
||
"oxidants"
|
||
],
|
||
"target": "D",
|
||
"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 sciq \
|
||
--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=['sciq'],
|
||
limit=10, # 正式评估时请删除此行
|
||
)
|
||
|
||
run_task(task_cfg=task_cfg)
|
||
``` |