def tk_loop
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
699def tk_loop(k: Nat, +e: U32, +l: U32, +t32: U32, +kv: U32, +hd32: U32, +kvw: U32, s: TG.O2) -> TG.O2:
700 match k s:
701 case 0n r: r
702 case 1n+q TG.O2{a, o}: tk_loop(q, (e + l : U32), l, t32, kv, hd32, kvw, tk_put(e, o, Array.get(F32, a, tk_src(e, t32, kv, hd32, kvw))))