paddle本地数据按内部格式加载

不能浅拷贝，是支持直接赋值的，paddle内部集成的数据集是支持，本地数据按内部格式直接加载的！

My guess is that you used code like dev_ds=train_ds, which is a copy instead of deepcopy. So there is only one dataset in memory with two names train_ds and dev_ds. When the program start to process dev_ds, it is actually processing train_ds again. That is why convert_example raise a exception since the data inside has already changed. The code below works fine.