Kaihua Liang

> dnn/src/cuda/multi_head_attn/opr_impl.cpp：can_use_mha_cudnn() 函数要完善下 好的，是指要让开启新功能时暂时不要用cudnn吗？

已解决相关问题，请检查。 @Ysllllll

> In the original implementation, 90 degree rotation is used as a data augmentation on omniglot. I can't find such preprocessing in this implementation. maybe this is the reason? @dragen1860...

It had been discussed in #32. Seems like the author have ignored this repository.....