Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Chinese-BERT-wwm基础上做预训练的方式 #17

Closed
brightmart opened this issue Jul 3, 2019 · 7 comments
Closed

Chinese-BERT-wwm基础上做预训练的方式 #17

brightmart opened this issue Jul 3, 2019 · 7 comments

Comments

@brightmart
Copy link

如何使用Chinese-BERT-wwm,在特定领域上再做预训练即操作方式?

@ymcui
Copy link
Owner

ymcui commented Jul 3, 2019

很遗憾,目前我们暂无计划开源代码,具体参考 #10 #13
事实上实现起来并不难,只需要修改数据生成的部分。
另外就是学习率不能设置的太大,这一点参考谷歌官方的Tips:https://github.com/google-research/bert#pre-training-tips-and-caveats

@ymcui
Copy link
Owner

ymcui commented Jul 5, 2019

reopen if necessary

@ymcui ymcui closed this as completed Jul 5, 2019
@lshowway
Copy link

很遗憾,目前我们暂无计划开源代码,具体参考 #10 #13
事实上实现起来并不难,只需要修改数据生成的部分。
另外就是学习率不能设置的太大,这一点参考谷歌官方的Tips:https://github.com/google-research/bert#pre-training-tips-and-caveats

我喜欢弹琵琶,经过wwm处理变成,我喜欢[mask][mask];那计算MLM loss的时候,真实label是琵琶(长度为1), 还是琵 琶(长度为2)?就是计算loss的时候“琵琶”是按两个字处理还是一个词?

@ymcui
Copy link
Owner

ymcui commented Nov 18, 2019

@lshowway 两个,wwm只改变输入mask,不改变loss计算。原来按字现在还是字。

@lshowway
Copy link

@lshowway 两个,wwm只改变输入mask,不改变loss计算。原来按字现在还是字。

谢谢您的回复。请问n-gram mask是不是与上述处理一致,原来按字现在还是字?期待您的回复。

@lshowway
Copy link

@lshowway 两个,wwm只改变输入mask,不改变loss计算。原来按字现在还是字。

谢谢您的回复。请问n-gram mask是不是与上述处理一致,原来按字现在还是字?那么n-gram mask可以以n-gram为单位么,将n-gram看成一个整体被预测,计算的也是probability over n-gram vocab size而不再是原来的probability over vocab size?
期待您的回复。

@lzy1012
Copy link

lzy1012 commented Dec 11, 2019

数据可以发下地址吗?汉语自然语言处理-BERT的解读语言模型预训练-实践应用-transformer模型(二)-语料预处理-情感分析分类

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants