def cx_t1
import bendtorch-gpu@1.3.0.0/main.bend as BendtorchGpuimport Base
1578# the host's part over st, row by row
1579def cx_t1(ok: Bool, +i: U32, +loss: F32, r: Array<F32> & F32) -> LS:
1580 match r:
1581 case Tuple{a, +xm}: ce_t2(ok, loss, xm, Array.get(F32, a, ((i * 3 : U32) + 2 : U32)))