Highway Transformer

Highway Transformer ：自门增强型自我关注网络

该仓库是使用Self-Dependency Unit的 Transformer-XL 的演示代码。这项工作与 Gating-enhanced Transformer 变体密切相关，例如 Google 的 Switch Transformer。

柴业坤等。等人。 ， Highway Transformer ：自门增强型自我关注网络 (ACL 2020)

要求

• PyTorch >= 1.1.0
• TensorboardX >= 1.8
• 张量板 >= 1.14
• 4 个 GPU ，每个8GB内存，用于运行 12 层 Transformer-XL

资料下载

bash getdata.sh

运行 6 层 Transformer-XL

 cd pytorch/xl_L6_scripts && bash < script-name > .sh train --work_dir " PATH_TO_WORK_DIR " 

可视化您的结果

 cd XL-L6-results && tensorboard --logdir=.

结果

• 不同模型设置的线图，其中最上面的线（红色）是基线模型（即原始 Transformer-XL）。
• 添加自依赖单元（参见底部两条曲线）后，很明显， Highway Transformer加快了训练和评估过程中的收敛过程。

训练业务流程	训练损失

评估bpc	评估损失

引文

对于学术背景下的归属，请将本作品引用为：

Highway Transformer: Self-Gating Enhanced Self-Attentive Networks", author = "Chai, Yekun and Jin, Shuo and Hou, Xinwen", booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics", month = jul, year = "2020", publisher = "Association for Computational Linguistics", url = "https://www.aclweb.org/anthology/2020.acl-main.616", pages = "6887--6900" }">

 @inproceedings{chai-etal-2020-highway,
    title = " Highway Transformer : Self-Gating Enhanced Self-Attentive Networks",
    author = "Chai, Yekun  and
      Jin, Shuo  and
      Hou, Xinwen",
    booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2020",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.acl-main.616",
    pages = "6887--6900"
}