Created by: chenwhql

Required（必填, multiple choices, two at most）

• PR type（PR 类型） is ( B ): A. New features（新功能）---------------- D. Performance optimization（性能优化） B. Bug fixes（问题修复）------------------ E. Breaking changes（向后不兼容的改变） C. Function optimization（功能优化）------F. Others（其它）

• PR changes（改动点）is ( D ): A. OPs（operators）---------------------- C. Docs（文档） B. APIs（接口）--------------------------- D. Others（其它）

• Use one sentence to describe what this PR does.（简述本次PR的目的和改动）

This PR fix random using problem in dy2static unittest test_bert, fix the problem in https://github.com/PaddlePaddle/Paddle/pull/24692.

Optional（选填, If None, please delete it）

• Describe what this PR does in detail. If this PR fixes an issue, please give the issue id.

test_bert data reader shows different generate result with the same seed, this is very strange, so we suspect that the DataLoader made changes to the input data.

But after testing, I found that these are the problems of numpy.random and python random, not the problem of DataLoader

1. numpy.random.seed is not thread-safe, but fluid.DataLoader use child thread to speed up data reading, the seed may be different in different thread even if you set the seed by np.random.seed. related doc:

• Numpy: random seed and multithreading causes differing results
• Differences between numpy.random and random.random in Python

2. python random has bug in python2, we should avoid using it in unittest. related doc:

• random.randint shows different output in Python 2.x and Python 3.x with same seed

3. for this case, we can use numpy.random.RandomState

• If you modified docs, please make sure that both Chinese and English docs were modified and provide a preview screenshot. （文档必填）

• Please write down other information you want to tell reviewers.