4 ms·Sequential Attention Making models leaner and fast without sacrificing accuracy2 points by binsquare 8mo ago