Symbol a = Symbol::Variable("a");
    Symbol b = Symbol::Variable("b");
    b = elemwise_add("added_b", a, b);

    vector<mx_float> vA, vB;
    vA.push_back(3.);
    vB.push_back(5.);
    auto tmpA = NDArray(vA.data(), Shape(1), ctx_cpu);
    auto tmpB = NDArray(vB.data(), Shape(1), ctx_cpu);
    args_map["a"] = tmpA;
    args_map["b"] = tmpB;
    Executor *exe = b.SimpleBind(ctx_cpu, args_map);
    exe->Forward(false);

    const auto &out = exe->outputs;
    NDArray out_cpu = out[0].Copy(ctx_cpu);
    NDArray::WaitAll();
    const mx_float *dptr_out = out_cpu.GetData();
    fprintf(stderr, "b = %f\n", dptr_out[0]);