search

brightmart / albert_zh

A LITE BERT FOR SELF-SUPERVISED LEARNING OF LANGUAGE REPRESENTATIONS, 海量中文预训练ALBERT模型
https://arxiv.org/pdf/1909.11942.pdf
3.94k stars 753 forks source link

加载使用albert base正常，加载albert tiny or small报错 #132

Closed sataliulan closed 3 years ago

sataliulan commented 4 years ago

ValueError: Asked to get output at node 11, but the layer has only 4 inbound nodes. 如上所示，我利用bert4keras加载albert base是正常的，换成albert的其他版本就不行了，不知道哪位大佬遇到过么？

Geekzhangwei commented 4 years ago

那是因为你的bert_layers设置的不是4，改为4就可以了。