Struct Muon 

pub struct Muon {
    pub learning_rate: LearningRate,
    pub momentum: f64,
    pub weight_decay: f64,
    pub nesterov: bool,
    pub ns_steps: usize,
    /* private fields */
}

Muon, as MLX’s optimizers.Muon: momentum (Nesterov by default) whose update is orthogonalised by Newton–Schulz iterations for parameters of two or more dimensions.

Fields

learning_rate: LearningRate

The learning rate.

momentum: f64

The momentum (MLX’s default 0.95).

weight_decay: f64

Weight decay added to the gradient (MLX’s default 0.01).

nesterov: bool

Nesterov momentum (MLX’s default).

ns_steps: usize

Newton–Schulz iterations (MLX’s default 5).

Implementations

impl Muon

pub fn new(learning_rate: impl Into<LearningRate>) -> Self

Muon with MLX’s defaults.

Trait Implementations

impl Clone for Muon

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Muon

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Optimizer for Muon

fn apply( &mut self, parameters: &[Array], gradients: &[Array], ) -> Result<Vec<Array>>

One step: the updated parameters.

fn update<M: Module>( &mut self, model: &mut M, gradients: &[Array], ) -> Result<()>

Applies one step to model: gradients are in the order of Module::trainable_parameters.