Andy Wong

这个模型不管标注，你改成和我用的数据一样的格式就行了，其他的跟readme里一样

训练时间足够长就会关注到少量非O带来的小loss，至于更好的方法等待大家去探索，评价时并不需要考虑到标签维度，只考虑实体类型和对应的各部分偏移量是否与gold data一一对应

readme里的Evaluation部分，有其他问题再看看其他closed或者已经回答过的issues。代码不难，自己看一遍也差不多懂了

> Hi, > I trained the model with the uploaded code, but couldn't reproduce the performance on the paper. > Is the settings in the code, such as random seed,...

> and I doubt the realness of the paper > > Can you provide pretrained model which can give the results reported in the paper. Could you upload your results?

> > > and I doubt the realness of the paper > > > Can you provide pretrained model which can give the results reported in the paper. > >...