About MHSA PROBLEMS.

[jiaaihhy]

Hello.
In your paper, you said that you use multi head self attention, and this head num is four.
But in this code, I only see 1 head in mhsa module.
And another question is why use this add to get postion encoding.
Looking forward for your reply!
Thank you!

[leaderj1001]

This repo is not official.
Also I will apply multi head soon.
Thanks.