Crate mlx_optim 

Optimizers for MLX training.

Tensors are immutable graph node handles — optimizers create new tensors for updated parameters rather than mutating in place.

Structs

AdamW

AdamW optimizer (Adam with decoupled weight decay).

CosineAnnealingLR

Cosine annealing learning rate scheduler.

Sgd

Stochastic Gradient Descent with optional momentum.

StepLR

Step-decay learning rate scheduler.

Traits

LrScheduler

Trait for learning rate schedulers.

Optimizer

Optimizer trait: apply one step, returning updated parameters.