数据

[dair-iitd]

数据

[rathorevipul28]

Yes, this code works even when there is only one relation between entities e1 and e2, and it can handle the 'NA' class as well!

[wangxinzhe123]

Because the maximum number of tokens allowed to be entered by bert is 512, but the length of the connected statements in my bag exceeds 512. If I don't want to discard the data after 512 bits, is there a solution to this situation?
(因为bert允许输入的最大tokens数目为512，但我bag中的语句连接之后长度超过512，如果我不想舍弃512位之后的数据，请问这种情况有解决办法吗？)

[rathorevipul28]

Even in our case, the length of combined sentences exceeded 512. But we showed in our paper that this doesn't hurt the performance compared to the baseline models.
But you can try other encoders like AllenNLP's longformer (https://github.com/allenai/longformer) supporting upto 16k tokens in the sequence.
Also checkout Transformer XL (https://github.com/kimiyoung/transformer-xl).

[wangxinzhe123]

好的，非常感谢