Alex Lin

python3.8，环境同requirements.txt 当使用多个gpu训练时候，会报错 RuntimeError: Input tensor at index 2 has invalid shape [2, 10, 12, 1024, 64], but expected [2, 11, 12, 1024, 64] 出错行显示在模型推理那行 outputs = model.forward(input_ids=batch_inputs, labels=batch_inputs) 这就很奇怪了，明明input_ids和labels完全是一个东西，却会报shape invalid...

这个东西不适合工业化应用

1

1. 存在内存泄漏，作为web服务时会造成内存一直增长，排查好久发现是lac分词的问题 2. 存在擅自更改logging的行为，也是排查好久 没话说了

time cost: msr 87.1 pku 35.1 other 93.6 效果 msr pku other cor 86.8 85.6 87.5 rec 86.0 80.7 84.6 f1 86.4 83.1 86.0

After reading some issues, I realized that it would cost a lot of time to train and take a heavy resouce to build a model on my own env. So...