int64_t FlattenedOutDimensionSize() const {
    if (ShapedType::isDynamicShape(out_dimensions_.sizes)) {
      return ShapedType::kDynamic;
    }
    return std::accumulate(out_dimensions_.sizes.begin(),
                           out_dimensions_.sizes.end(), 1,
                           std::multiplies<int64_t>());
  }

 private:
  DimensionVector batch_dimensions_;

  int64_t FlattenedOutDimensionSize() const {
    if (ShapedType::isDynamicShape(out_dimensions_.sizes)) {
      return ShapedType::kDynamic;
    }
    return std::accumulate(out_dimensions_.sizes.begin(),
                           out_dimensions_.sizes.end(), 1,
                           std::multiplies<int64_t>());
  }

 private:
  DimensionVector batch_dimensions_;

                          ArrayRef<int64_t> output_shape, int num_dimensions,
                          int output_axis, std::vector<uint64_t>* input_indices,
                          std::vector<Attribute>* new_values) const {
    // Refer to the implementation of `Transpose` function in
    // tensorflow/lite/kernels/internal/reference/reference_ops.h
    assert(output_axis < num_dimensions);
    const int input_axis = perm[output_axis];

  SmallVector<int64_t> output_shape;
  ArrayRef<int64_t> input_shape = input_tensor.getType().getShape();
  auto attr_iter = perm_tensor.getValues<IntegerAttr>();
  const int num_dimensions = input_tensor.getType().getRank();
  for (int i = 0; i < num_dimensions; ++i) {
    perms.push_back(attr_iter[i].getInt());
    output_shape.push_back(input_shape[perms[i]]);
  }

  typedef StablehloReduceOptionsBuilder Builder;
  enum FlatBuffersVTableOffset FLATBUFFERS_VTABLE_UNDERLYING_TYPE {
    VT_DIMENSIONS = 4,
    VT_BODY_SUBGRAPH_INDEX = 6
  };
  const ::flatbuffers::Vector<int64_t> *dimensions() const {
    return GetPointer<const ::flatbuffers::Vector<int64_t> *>(VT_DIMENSIONS);
  }
  int32_t body_subgraph_index() const {
    return GetField<int32_t>(VT_BODY_SUBGRAPH_INDEX, 0);
  }