def Q8.matmul_nt

bendtorch@1.3.0.0 · dtype.bend:497

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

497# y = x · Wᵀ; x and W come back
498def Q8.matmul_nt(+n: Nat, +k: Nat, +m: Nat, x: TA.Mat<n, k>, w: Q8<m, k>) -> Q8Mul<n, k, m>:
499  match x w:
500    case TA.Mat{xa} Q8{q, s}:
501      q8_mm_fin(n, k, m, q8_rowsm(n, 0, m, k, U32.from_nat(k), U32.from_nat(m), QS{xa, q, s, T.zeros_a(Nat.mul(n, m))}))