1 paper
Nicola Zanarini, Niccolò Ferrari, Evelina Lamma
We investigate whether the Feed-Forward Network (FFN) sublayer in a decoder-only transformer can be replaced by an explicit learned memory graph while preserving the surrounding au…