silu(gate(x))·up(x) for 1..4 activation rows when gate and up share their axis-1 geometry but not their bit width. Borrows x and the weights; returns owned lazy output with x's leading dimensions.
silu(gate(x))·up(x) for 1..4 activation rows when gate and up share their axis-1 geometry but not their bit width. Borrows x and the weights; returns owned lazy output with x's leading dimensions.