def Attention

bendtorch-fast@1.3.0.0 · main.bend:336

import Base
import bend-ml-tensor-array@0.1.18.0/main.bend as TA
import bendtorch@1.3.0.0/main.bend as T

336def Attention(+n: Nat, +w: Nat, +kvw: Nat, +b: Nat, +t: Nat, +kv: Nat, +g: Nat, +hd: Nat, +sc: F32, -e1: {n == Nat.mul(b, t) : Nat}, -e2: {w == Nat.mul(Nat.mul(kv, g), hd) : Nat}, -e3: {kvw == Nat.mul(kv, hd) : Nat}) -> T.Layer<Unit, T.Two<TA.Mat<n, w>, T.Two<TA.Mat<n, kvw>, TA.Mat<n, kvw>>>, TA.Mat<n, w>, T.AttnS>:
337  T.Attention(n, w, kvw, b, t, kv, g, hd, sc, e1, e2, e3)