You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
taoht b2c1cd7722 update 1 year ago
..
__init__.py update 1 year ago
dataset.py update 1 year ago
generate.py update 1 year ago
pangu_alpha.py update 1 year ago
pangu_alpha_config.py update 1 year ago
pangu_alpha_wrapcell.py update 1 year ago
preprocess.py update 1 year ago
serialization.py update 1 year ago
tokenization_jieba.py update 1 year ago
utils.py update 1 year ago

mPanGu-α-53是首个以中文为中心的多语言&机器翻译模型,在一带一路沿线66个国家53种语种上进行预训练和单双语混合增量训练,单模型支持一带一路53个语种任两语种间的互译,对比WMT2021多语言任务赛道No.1在”中外“100个方向上平均BLEU值提升0.354,支持在NPU/GPU上基于MindSpore分布式训练(最少8卡)、推理(全精度/FP16,1卡)和多语言任务的迁移学习。

Text Python

Apache-2.0

Contributors (2)