Some questions about the model and dataset

您好，我在看了源码后有些问题，希望作者能帮忙解答下，感谢。
1、在Attentive Module的FFN层里面，权重初始化用了orthogonal_initializer()，请问为什么用到这种初始化？
2、 FFN层的隐藏层维度大小和输入维度相同，请问是否有测试过其他维度？
3、请问ubuntu dataset里面的word_embedding.pkl文件，是自己训练出来的吗？如果是，请问有没有相应的代码？
希望作者可以回答下，谢谢了哈~

Start with the Attentive Module's FFN and the Ubuntu dataset assets mentioned in the issue. Trace the orthogonal initializer, compare the stated hidden and input dimensions, and inspect word_embedding.pkl for its provenance. Done means the initialization rationale, dimension experiments, and embedding-generation process are documented or answered.

Written by the indexing model from the issue text.