resolve #1878 !1879
Created by: ZeyuChen
Resolve https://github.com/PaddlePaddle/models/issues/1878
修复PaddleNLP中文本分类训练的bug。
- 命令行传GRU不生效
- train函数中的batch_size为冗余参数,batch size由reader控制
- GRU默认的embbeding layer初始化学习率过大,导致不收敛,已调整到与LSTM一致
Created by: ZeyuChen
Resolve https://github.com/PaddlePaddle/models/issues/1878
修复PaddleNLP中文本分类训练的bug。