Evaluation procedure？

### Description

I wittiness there are lots of evaluation benchmark in paper, how can I reproduce the evalution procedure? I want to    compare the result using fintuned model

### Reproduction

asdsad

### Logs

```shell

```

### Environment Information

asdasd

### Known Issue

- [x] The issue hasn't been already addressed in Documentation, Issues, and Discussions.