Is this model suitable for chinese?

[Waste-Wood]

I have seen some files like "char_vocab_chinese.txt", so is there a chinese pretrained model?

[henryhust]

That 's ok, maybe you need to train a model of your own

[Waste-Wood]

That 's ok, maybe you need to train a model of your own

Except the GloVe embedding that I should replace with my own chinese embedding, is there any file that I should replace with my own file? And I have seen training data of chinese is included in these files!

[mjj1094]

Hope that someone has trained this model in Chinese and would like to share his github link here

[viviqi]

I am trying train the model with Chinese dataset from Ontonote5.0, but it seems that the model can not converge, if anyone has done the same work, advice will be welcome.

[mjj1094]

Thanks for your reply！ I just referenced https://github.com/mandarjoshi90/coref for the conversion of Chinese data，maybe you can try it. On 06/30/2020 13:59, zhuqi wrote: I am trying train the model with Chinese dataset from Ontonote5.0, but it seems that the model can not converge, if anyone has done the same work, advice will be welcome. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[viviqi]

Thanks for your advice, yes I am trying the bert-coref model now and the performance is not good yet, anyway, still finetune the model. And if you are training it as well, maybe we can share our results.

[mjj1094]

I am training(not finished) the model of Chinese-coref-bert from https://github.com/mandarjoshi90/coref which is an extension of the e2e-coref model. If you want a further communication,you can contact me by my e-mail:[email protected]. On 07/01/2020 16:43, zhuqi wrote: Thanks for your advice, yes I am trying the bert-coref model now and the performance is not good yet, anyway, still finetune the model. And if you are training it as well, maybe we can share our results. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.