Skip to content

Latest commit

 

History

History
9 lines (9 loc) · 602 Bytes

README.md

File metadata and controls

9 lines (9 loc) · 602 Bytes

paddle_mbart

论文复现-Multilingual Denoising Pre-training for Neural Machine Translation Pretrained language model 论文名称: Multilingual Denoising Pre-training for Neural Machine Translation 数据集: 2010-2013 TED Zh-En测试集 WMT16 EN-RO WMT16 Ro-En 验收标准: 1. 复现mbart模型(参考论文和实现链接) 2. 完成模型权重从pytorch到paddle的转换代码,转换1个预训练权重(“facebook/mbart-large-cc25”) 3. "facebook/mbart-large-cc25"模型指标: 在WMT16 EN-RO测试集上BLEU scores=37.7, 在WMT16 Ro-En测试集上BLEU scores=37.8