关于不能用RobertaModel引用

[Jackthebighead]

您好，在readme中的快速加载有这样一句：

注意：本目录中的所有模型均使用BertTokenizer以及BertModel加载，请勿使用RobertaTokenizer/RobertaModel！

我从huggingface官网下下来了roberta-chinese-wwm的pytorch_model.bin，tokenizer.json和vocab，再想在代码使用.from_pretrained()方法引用，即不是采用快速加载的情况，请问这样的话是用RobertaModel引用还是BertModel引用呢？我尝试用BertModel引用，但是提示:

You are using a model of type roberta to instantiate a model of type bert. This is not supported for all configurations of models and can yield errors.`

还请解答一下，非常感谢！

[xiaoguzai]

如果你的from_pretrained放的是下载的路径的话，就是bertmodel引用robertamodel的预训练参数 不过这不影响使用，因为bertmodel和robertamodel的微调部分代码是一样的

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[liushunge]

您的邮件已收到，谢谢。

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[Jackthebighead]

如果你的from_pretrained放的是下载的路径的话，就是bertmodel引用robertamodel的预训练参数 不过这不影响使用，因为bertmodel和robertamodel的微调部分代码是一样的

明白了，感谢。

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale[bot]]

Closing the issue, since no updates observed. Feel free to re-open if you need any further assistance.

[gaodanao]

老哥，你好，请问这个问题不需要解决吗 因为我在使用dietilbert也遇到同样的问题