NLP/nlp_xiaojiang

History

yongzhuo 729f8ee20d fix readme.md of augment		2021-09-23 13:44:23 +08:00
..
__init__.py	add augment-nmt and augment-simbert	2021-09-23 11:34:12 +08:00
enhance_roformer.py	fix readme.md of augment	2021-09-23 13:44:23 +08:00
enhance_simbert.py	fix readme.md of augment	2021-09-23 13:44:23 +08:00
README.md	fix readme.md of augment	2021-09-23 13:44:23 +08:00
requestments.txt	add augment-nmt and augment-simbert	2021-09-23 11:34:12 +08:00
tet_keras.py	fix readme.md of augment	2021-09-23 13:44:23 +08:00

README.md

Augment Simbert

预训练模型-UMILM

使用BERT(UNILM)的生成能力, 即BERT的NSP句对任务 - simbert(bert + unilm + adv): https://github.com/ZhuiyiTechnology/simbert - simbert: 鱼与熊掌兼得：融合检索和生成的SimBERT模型 - roformer-sim: https://github.com/ZhuiyiTechnology/roformer-sim - simbert-v2(roformer + unilm + adv + bart + distill): SimBERTv2来了！融合检索和生成的RoFormer-Sim模型

备注

效果还是比较好的, 可以生成多个相似句子, 但是生成式的模型一般都比较慢。

相比simbert, roformer-sim效果要好些