Geong
Geong
Yes, please just ignore it. I will fix it soon.
Indeed, due to the GitHub file size limit, the ckpt file was not uploaded. I will commit an update with a download link in the near future. You can also...
很抱歉，由于项目知识产权原因，暂时不能开源完整的训练代码。事实上，代码已经包含了所有训练所需的模块及数据预处理方法。只需要在training_step函数中像论文所述的那样将三个loss相加就可以了，无需额外的训练技巧。具体实现代码可以参考BTTR和CoMER。
Indeed, this file is missing, but in reality, the result has already been calculated during inference and there is no need to call offline_ eval module. You can just ignore...
您好，很高兴您能关注到我们的工作。 建议您先跑通[BTTR](https://github.com/Green-Wood/BTTR)，然后只需要加入论文中提到的多行mask和loss就可以了，没有额外的trick。