gro?moc?ten?ude?vog??g?il?mi?orp??n!.&a&0&b-ekhgnark--nx?c-iehsrgev--nx?g-lksedlig--nx?k-negnanvk--nx??1&p-nedragy--nx?q-&asierrs--nx?grebsnt--nx?lado-rs--nx?n&egnidl--nx?orf-rs--nx??regnayh--nx?ssofenh--nx??r-datsgrt--nx?s-ladrjts--nx?v-y&senner--nx?vrejks--nx???3g-datsobegh--nx?4&5-&dnaleprj--nx?goksnerl--nx?tednalyh--nx??6-neladnjm--nx?s-&antouvachb--nx?impouvtalm--nx??y-&agrjnevvad--nx?ikhvlaraeb--nx???7k-antouvacchb--nx?8&k-rekie-erv--nx?l-ladrua-rs--nx?m-darehsdrk--nx??a!.sg??bct-eimeuvejs...

// `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