pre.py 改进建议

**Summary[问题简述（一句话）]**
使用 `pre.py` 时如何暂停以及开始

**Env & To Reproduce[复现与环境]**
各依赖环境正常

使用 `pre.py` 开始训练后，无法停止
按 `Ctrl + C` 后报错，但计算未停止
看了一下 `pre.py` 源码，应该是 `multiprocessing` 的问题，其进程实例需人为停止
可以考虑使用 ：
`p.terminate()`
`p.join()`

Start with pre.py and reproduce the training run described in the issue, then inspect how its multiprocessing instances respond to Ctrl+C. Verify that stopping the run also stops and joins the worker processes, and confirm that training no longer continues after the interruption.

Written by the indexing model from the issue text.