1 paper · 1 filter
O. Duranthon, P. Marion, C. Boyer +2
Large language models rely on attention mechanisms with a softmax activation. Yet the dominance of softmax over alternatives (e.g., component-wise or linear) remains poorly underst…