I am working on an implementation for the Resize operator for axon_onnx here: Support for `op_type` "Resize" by jnnks · Pull Request #35 · elixir-nx/axon_onnx · GitHub and need to calculate the output shape of the Tensor inside a defn context.

Consider the following snippet:

defmodule Sample do
  import Nx.Defn

  defn calc_output_size(input, scales) do
    # take inputs shape and multiply with scales
    #   input: Nx.Tensor<f32[1][1][2][4]>
    #   scales: Nx.Tensor<f32[4]>
    # return tuple
    output_size = input
      |> Nx.shape()
      |> Tuple.to_list()
      |> Nx.tensor()
      |> Nx.multiply(scales)
      |> Nx.to_flat_list()
      |> List.to_tuple()
  end
end

The call to Nx.to_flat_list fails during compile-\runtime.
How can I get the values out of the tensor and into a tuple?