def lf_b1_f
import bendtorch-gpu@1.3.0.0/main.bend as BendtorchGpuimport Base
import bend-ml-tensor-gpu@0.1.5.0/main.bend as TG
1061# x1 = q, x2 = kᵀ, x3 = vᵀ, x4 = dO; outputs P and G-dP (made here, written)
1062def lf_b1_f(+t: U32, +l: U32, +c: EAC, x1: Array<F32>, x2: Array<F32>, x3: Array<F32>, x4: Array<F32>, o: TG.O2) -> TG.O2:
1063 match o:
1064 case TG.O2{p, g}: b1_o(b1_loop_f(TG.share(t, eac_n(c), l), t, l, c, S6{x1, x2, x3, x4, p, g}))