gated Transformer
1.0.0
具有稳健的deNo序列对序列建模的封闭术验证的变压器模型。它使用门控单元来检测和纠正文本数据的噪声,并从生成解码器中生成掉落的目标。

在这项工作中,我们对三个任务进行实验 -
封闭式转换器在 -
$ pip install -r requirements.txt
预处理的变压器序列到序列模型(例如 - BART,T5)
$ cd ./drive/MyDrive/gated-denoise/ && python train.py --train_file ./data.csv
--model_path ./model/ --model_type bart --pretrained_encoder_path "facebook/bart-base"
--mask_gate --copy_gate --generate_gate --skip_gate
--epochs 15
变压器编码器模型(例如-Bert2bert)
$ cd ./drive/MyDrive/gated-denoise/ && python train.py --train_file ./data.csv
--model_path ./model/ --model_type seq2seq
--pretrained_encoder_path "bert-base-uncased" --pretrained_decoder_path "bert-base-uncased"
--mask_gate --copy_gate --generate_gate --skip_gate
--epochs 15
$ cd ./drive/MyDrive/gated-denoise/ && python predict.py --data_file ./data.csv
--model_path ./model/ --model_type bart --pretrained_encoder_path "facebook/bart-base"
--mask_gate --copy_gate --generate_gate --skip_gate
如果您发现此存储库有用,请引用我们的论文:
@inproceedings {,
author = { Ayan Sengupta and
Amit Kumar and
Sourabh Kumar Bhattacharjee and
Suman Roy } ,
title = { Gated Transformer for Robust De-noised Sequence-to-Sequence Modelling } ,
booktitle = { } ,
publisher = { } ,
year = { } ,
url = { } ,
doi = { } ,
}