Struct DiscriminativeFineTuning 

pub struct DiscriminativeFineTuning {
    pub base_lr: f64,
    pub decay_factor: f64,
    /* private fields */
}

Discriminative fine-tuning: use different learning rates for different layers.

Typically, earlier layers use smaller learning rates than later layers.

Fields

base_lr: f64

Base learning rate for the last layer.

decay_factor: f64

Learning rate decay factor (each earlier layer uses lr * decay_factor).

Implementations

impl DiscriminativeFineTuning

pub fn new(base_lr: f64, decay_factor: f64) -> TrainResult<Self>

Create a new discriminative fine-tuning configuration.

Arguments

• base_lr - Learning rate for the last layer
• decay_factor - Decay factor for earlier layers (e.g., 0.5 means half LR)

pub fn compute_layer_lrs(&mut self, layer_order: &[String])

Compute learning rates for all layers.

Arguments

• layer_order - Layer names ordered from bottom to top

pub fn get_layer_lr(&self, layer_name: &str) -> f64

Get the learning rate for a specific layer.

Arguments

• layer_name - Name of the layer

Returns

Learning rate for the layer, or base_lr if not found

pub fn layer_lrs(&self) -> &HashMap<String, f64>

Get all layer learning rates.

Trait Implementations

impl Clone for DiscriminativeFineTuning

fn clone(&self) -> DiscriminativeFineTuning

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DiscriminativeFineTuning

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

Formats the value using the given formatter.