https://hgpu.org/?p=17130
Unfolding and Shrinking Neural Machine Translation Ensembles