README.md

## MATH-500数据集
- https://www.modelscope.cn/datasets/AI-ModelScope/MATH-500

## vllm 脚本修改

- serve.py 修改
```
mv /usr/local/lib/python3.10/dist-packages/vllm/benchmarks/serve.py /usr/local/lib/python3.10/dist-packages/vllm/benchmarks/serve.py.bak

cp ./utils/vllm-benchmarks/serve.py /usr/local/lib/python3.10/dist-packages/vllm/benchmarks/
```

- datasets.py修改
```
mv /usr/local/lib/python3.10/dist-packages/vllm/benchmarks/datasets.py /usr/local/lib/python3.10/dist-packages/vllm/benchmarks/datasets.py.bak

cp ./utils/vllm-benchmarks/datasets.py /usr/local/lib/python3.10/dist-packages/vllm/benchmarks/
```

## evalscope 修改
- evaluator.py
```
cp /usr/local/lib/python3.10/dist-packages/evalscope/evaluator/evaluator.py /usr/local/lib/python3.10/dist-packages/evalscope/evaluator/evaluator.py.bak

cp ./utils/evalscope/evaluator.py /usr/local/lib/python3.10/dist-packages/evalscope/evaluator/
```

## 启动vllm服务
- bash vllm_serve.sh

## 性能和精度结果保存

- bash run_benchmarks.sh

## 数据集转换
```
cd tools
bash run_convert.sh
```

## 跑精度
```
cd tools
bash evalscope_test.sh
```