helloword12345678 Goto Github PK

chineseasr

examples

fastbidaf

lhotse

neural-summ-cnndm-pytorch

记录真正的copy trick：是将source的oov也记录下来，假设词典1000， 那么oov的id就是1000，1001.。 然后维护两个encoder的输入x，一个是有unk的x，一个是把unk的id替换成oov id的x，copy的计算因为只需要跟hidden算，所以oov也有attention。然后在最后output的时候把oov拼接到1000的pred_y上，这样id是1001的词也就被copy出来了。 提升果然非常非常大。

nq_model

multi-spans extraction model based on BERT

pan-baidu-download

百度网盘下载脚本

recurrentshop

Framework for building complex recurrent neural networks with Keras

rocketqa

🚀 RocketQA, dense retrieval for information retrieval and question answering, including both Chinese and English state-of-the-art models.

you-only-speak-once

Deep Learning - one shot learning for speaker recognition using Filter Banks