// exporting a full trace so other code can do backpropagation. Not thread-safe.
template <typename Gradient, typename BackwardFunction, typename TapeTensor>
class GradientTape {
 public:
  // If `persistent` is true, GradientTape will not eagerly delete backward
  // functions (and hence the tensors they keep alive). Instead, everything
  // is deleted in ~GradientTape. Persistent GradientTapes are useful when