// `dx` must be nullptr or have size `ny`.
// If `dx` is nullptr, the implementation will use dx of `OnesLike` for all
// shapes in `y`.
// The partial derivatives are returned in `dy`. `dy` should be allocated to
// size `nx`.
//
// Gradient nodes are automatically named under the "gradients/" prefix. To
// guarantee name uniqueness, subsequent calls to the same graph will

    // The expected graph looks like this if grad_inputs_provided.
    // If grad_inputs_provided is false, Const_0 will be a OnesLike op.
    //      ^             ^
    //    dy|           dx|        // MatMul Gradient Graph
    //      |             |
    //   MatMul_2      MatMul_1
    //   ^   ^          ^    ^
    //   |   |----------|    |
    //   |        ^          |