Module optim 

Optimizers for updating learnable parameters.

Per Tenet #4 (“Borrow-Safe Optimizers”): Optimizer::step takes &mut GradientStore and selectively drains only the gradients belonging to its registered parameters. Multiple optimizers can each drain their subset from the same store (e.g., GAN generator vs. discriminator).

Re-exports

pub use clip::clip_grad_norm_;

pub use scheduler::CosineAnnealingLR;

pub use scheduler::LRScheduler;

pub use scheduler::StepLR;

Modules

clip

Gradient clipping utilities.

scheduler

Learning rate schedulers.

Structs

Adam

AdamW

SGD

Traits

Optimizer

Trait for all optimizers.