}
    }
    TF_ShapeAndTypeList* output_shapes;
    TFE_InferShapes(op, input_shapes,
                    input_tensors.empty()
                        ? nullptr
                        : const_cast<TF_Tensor**>(input_tensors.data()),
                    /*input_tensors_as_shapes*/ nullptr,
                    /*input_resource_shapes_and_types*/ nullptr, &output_shapes,

//     OK to not have the inputs properly set in `op`. See `input_tensors`
//     if you want shape inference to consider the input tensors of the
//     op for shape inference.
//   - The types need not be set in `input_shapes` as it is not used.
//   - The number of `input_tensors` should be the same as the number of items
//     in `input_shapes`.
//

  });
  std::vector<Gradient*> in_grads;
  in_grads.reserve(input_tensors.size());
  for (int target_index = 0; target_index < input_tensors.size();
       ++target_index) {
    const auto current_grad =
        accumulated_gradients_.find(input_tensors[target_index].GetID());
    if (current_grad == accumulated_gradients_.end()) {
      if (IsDtypeTrainable(input_tensors[target_index].GetDType())) {

      for (float factor : tensor_property.derived_scale.factors) {
        scale *= factor;
      }
      spec->biases_params.emplace(
          index,
          std::make_pair(tensor_property.derived_scale.input_tensors,
                         GetUniformQuantizedTypeForBiasWithScale(scale)));
    }
  }
  return spec;
}

class ConvertSvdfStatsToQDQs : public ConvertOpStatsToQDQs<TFL::SVDFOp> {
 public: