MBart模型汇总#

下表汇总介绍了目前PaddleNLP支持的MBart模型对应预训练权重。 关于模型的具体细节可以参考对应链接。

Pretrained Weight

Language

Details of the model

mbart-large-cc25

English

12-layer, 1024-hidden, 12-heads, 1123M parameters. The mbart-large-cc25 model.

mbart-large-en-ro

English

12-layer, 768-hidden, 16-heads, 1123M parameters. The mbart-large rn-ro model.

mbart-large-50-one-to-many-mmt

English

12-layer, 1024-hidden, 16-heads, 1123M parameters. mbart-large-50-one-to-many-mmt model.

mbart-large-50-many-to-one-mmt

English

12-layer, 1024-hidden, 16-heads, 1123M parameters. mbart-large-50-many-to-one-mmt model.

mbart-large-50-many-to-many-mmt

English

12-layer, 1024-hidden, 16-heads, 1123M parameters. mbart-large-50-many-to-many-mmt model.