13274243a0
Keep K3 suite selection and report-schema scoring in bash, merge K3/vision dataset_args into dpv4 yamls, and pin EvalScope at 735d920ee911 with local patches. Co-authored-by: Cursor <cursoragent@cursor.com>
133 lines
4.4 KiB
Markdown
133 lines
4.4 KiB
Markdown
# BoolQ-Indic
|
||
|
||
|
||
## 概述
|
||
|
||
BoolQ-Indic 是将 BoolQ 是/否阅读理解基准测试翻译为 10 种印度语言及英语的版本，用于评估多语言段落理解能力。
|
||
|
||
## 任务描述
|
||
|
||
- **任务类型**：多语言是/否阅读理解
|
||
- **输入**：一段文章 + 一个用 11 种语言之一提出的是/否问题
|
||
- **输出**：`Yes` 或 `No`
|
||
- **语言**：孟加拉语、英语、古吉拉特语、印地语、卡纳达语、马拉雅拉姆语、马拉地语、奥里亚语、旁遮普语、泰米尔语、泰卢固语
|
||
|
||
## 评估说明
|
||
|
||
- 默认配置使用 **0-shot** 评估（验证集）
|
||
- 使用 `subset_list` 来评估特定语言（例如 `['hi', 'ta']`），或使用 `limit` 限制样本数量 —— 默认完整运行包含全部 11 种语言共 35,970 个样本
|
||
- 设置 `few_shot_num` > 0 可启用少样本提示；示例从 `train` 划分中抽取
|
||
- 所有语言均包含在单一数据集配置中；此适配器根据 `language` 字段重新格式化数据
|
||
|
||
## 属性
|
||
|
||
| 属性 | 值 |
|
||
|----------|-------|
|
||
| **基准测试名称** | `indic_boolq` |
|
||
| **数据集ID** | [sarvamai/boolq-indic](https://modelscope.cn/datasets/sarvamai/boolq-indic/summary) |
|
||
| **论文** | N/A |
|
||
| **标签** | `MCQ`, `MultiLingual`, `ReadingComprehension` |
|
||
| **指标** | `accuracy` |
|
||
| **默认样本数** | 0-shot |
|
||
| **评估划分** | `validation` |
|
||
| **训练划分** | `train` |
|
||
|
||
|
||
## 数据统计
|
||
|
||
| 指标 | 值 |
|
||
|--------|-------|
|
||
| 总样本数 | 35,970 |
|
||
| 提示词长度（平均） | 822.66 字符 |
|
||
| 提示词长度（最小/最大） | 275 / 5035 字符 |
|
||
|
||
**各子集统计数据：**
|
||
|
||
| 子集 | 样本数 | 提示词平均长度 | 提示词最小长度 | 提示词最大长度 |
|
||
|--------|---------|-------------|------------|------------|
|
||
| `bn` | 3,270 | 801.95 | 294 | 2308 |
|
||
| `en` | 3,270 | 814.26 | 292 | 5035 |
|
||
| `gu` | 3,270 | 793.37 | 283 | 2105 |
|
||
| `hi` | 3,270 | 818.47 | 297 | 3078 |
|
||
| `kn` | 3,270 | 833.99 | 275 | 2920 |
|
||
| `ml` | 3,270 | 869.99 | 294 | 3558 |
|
||
| `mr` | 3,270 | 806.68 | 289 | 2593 |
|
||
| `or` | 3,270 | 787.68 | 306 | 1482 |
|
||
| `pa` | 3,270 | 804.93 | 295 | 1975 |
|
||
| `ta` | 3,270 | 904.01 | 297 | 3570 |
|
||
| `te` | 3,270 | 813.95 | 284 | 3312 |
|
||
|
||
## 样例示例
|
||
|
||
**子集**: `bn`
|
||
|
||
```json
|
||
{
|
||
"input": [
|
||
{
|
||
"id": "0f04f0f7",
|
||
"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.\n\nসকল জৈববস্তুই কমপক্ষে এই ধাপগুলোর মধ্য দিয়ে যায়: এগুলো চা ... [TRUNCATED 1099 chars] ... বার্কলেতে ছয়টি পৃথক গবেষণা বিশ্লেষণ করার পর, একটি গবেষণায় উপসংহারে আসা গেছে যে, ভুট্টা থেকে ইথানল উৎপাদনে পেট্রোলিয়ামের ব্যবহার গ্যাসোলিন উৎপাদনের তুলনায় অনেক কম।\n\nQuestion: ইথানল উৎপাদনের চেয়ে তৈরিতে কি বেশি শক্তি লাগে??\n\nA) Yes\nB) No"
|
||
}
|
||
],
|
||
"choices": [
|
||
"Yes",
|
||
"No"
|
||
],
|
||
"target": "B",
|
||
"id": 0,
|
||
"group_id": 0,
|
||
"subset_key": "bn",
|
||
"metadata": {
|
||
"language": "Bengali"
|
||
}
|
||
}
|
||
```
|
||
|
||
*注：部分内容为显示目的已截断。*
|
||
|
||
## 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 indic_boolq \
|
||
--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=['indic_boolq'],
|
||
dataset_args={
|
||
'indic_boolq': {
|
||
# subset_list: ['bn', 'en', 'gu'] # 可选，用于评估特定子集
|
||
}
|
||
},
|
||
limit=10, # 正式评估时请删除此行
|
||
)
|
||
|
||
run_task(task_cfg=task_cfg)
|
||
```
|