//! Generic parameter optimization: update-rule traits, gradient clipping,
//! and the concrete optimizers built on them.
//!
//! Split by concern per `docs/CONVENTIONS.md`: `traits` is `Optimizer`/
//! `ScaledOptimizer` (the update-rule vocabulary) and `OptimizerBackend`/
//! `ValueClippingBackend` (the canonical execution profile a backend
//! implements to support them, plus the blanket impl over exact-op
//! descriptors that satisfies it); `group` is `ParameterGroup` and the two
//! private types its constructor and every optimizer's update loop build
//! on; `support` is the validation and update-staging logic shared across
//! `step`/`load_state_dict`; `clip` is the public gradient-clipping API;
//! `optimizers` is `SGD`, `AdamW` and `Adam` themselves.
pub use crateGradients;
pub use ;
pub use ParameterGroup;
pub use ;
pub use ;
pub use ;
pub use ;