wolverineq

Good to hear! and hope a multi-gpu training version will be added since training of attention-based model is quite time-consuming...