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>
129 lines
3.7 KiB
Markdown
129 lines
3.7 KiB
Markdown
# BhashaBench-V1 (Finance)
|
||
|
||
|
||
## 概述
|
||
|
||
BhashaBench-Finance 是 BhashaBench-Multi 金融领域的前身：一个领域特定的多项选择题基准测试，用于评估大语言模型（LLM）在金融领域的知识，涵盖英语和印地语。
|
||
|
||
## 任务描述
|
||
|
||
- **任务类型**：领域特定的多项选择题问答
|
||
- **输入**：一道包含4个选项的金融问题，语言为英语或印地语
|
||
- **输出**：正确答案对应的字母
|
||
- **语言**：英语、印地语
|
||
|
||
## 主要特点
|
||
|
||
- 每种语言包含5,600至17,000道题目，仅涵盖英语和印地语
|
||
- 作为 BhashaBench-Multi 的前身：领域相同，但语言覆盖范围更窄
|
||
- 每个领域对应一个独立的代码仓库，英语和印地语分别作为独立的配置项
|
||
|
||
## 评估说明
|
||
|
||
- 默认配置使用 **0-shot** 评估（仅提供 test 分割）
|
||
- 使用 `subset_list` 参数可评估单一语言（例如 `['Hindi']`）
|
||
- 需要访问此受限制的数据集：在 ModelScope（默认数据集中心）上，请先接受使用条款并确保已登录；或者将 `dataset_hub` 设置为 `huggingface`，并在 huggingface.co 上接受条款后使用 `HF_TOKEN`
|
||
- 如需在同一领域下获得更广泛的语言覆盖，请参考 `bhasha_bench_multi_finance`（涵盖22种印度语言，且无需授权）
|
||
|
||
## 属性
|
||
|
||
| 属性 | 值 |
|
||
|----------|-------|
|
||
| **基准测试名称** | `bhashabenchv1_finance` |
|
||
| **数据集ID** | [bharatgenai/BhashaBench-Finance](https://modelscope.cn/datasets/bharatgenai/BhashaBench-Finance/summary) |
|
||
| **论文** | N/A |
|
||
| **标签** | `Knowledge`, `MCQ`, `MultiLingual` |
|
||
| **指标** | `accuracy` |
|
||
| **默认Shots数** | 0-shot |
|
||
| **评估分割** | `test` |
|
||
|
||
|
||
## 数据统计
|
||
|
||
| 指标 | 值 |
|
||
|--------|-------|
|
||
| 总样本数 | 19,433 |
|
||
| 提示词长度（平均） | 612.82 字符 |
|
||
| 提示词长度（最小/最大） | 221 / 6665 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 |
|
||
|--------|---------|-------------|------------|------------|
|
||
| `English` | 13,451 | 663.98 | 223 | 6665 |
|
||
| `Hindi` | 5,982 | 497.79 | 221 | 3304 |
|
||
|
||
## 样例示例
|
||
|
||
**子集**: `English`
|
||
|
||
```json
|
||
{
|
||
"input": [
|
||
{
|
||
"id": "befc8699",
|
||
"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\nIn the following number series. One number is wrong. Find the wrong number of the series? 3, 4, 12, 38, 103, 228\n\nA) 103\nB) 12\nC) 38\nD) 228"
|
||
}
|
||
],
|
||
"choices": [
|
||
"103",
|
||
"12",
|
||
"38",
|
||
"228"
|
||
],
|
||
"target": "C",
|
||
"id": 0,
|
||
"group_id": 0,
|
||
"metadata": {
|
||
"language": "English",
|
||
"topic": "Quantitative Aptitude"
|
||
}
|
||
}
|
||
```
|
||
|
||
## 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 bhashabenchv1_finance \
|
||
--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=['bhashabenchv1_finance'],
|
||
dataset_args={
|
||
'bhashabenchv1_finance': {
|
||
# subset_list: ['English', 'Hindi'] # 可选，用于指定评估的子集
|
||
}
|
||
},
|
||
limit=10, # 正式评估时请删除此行
|
||
)
|
||
|
||
run_task(task_cfg=task_cfg)
|
||
```
|