Struct LoRAAdapter 

pub struct LoRAAdapter {
    pub lora_a: Tensor,
    pub lora_b: Tensor,
    pub scaling: f32,
    pub active: bool,
}

LoRA adapter containing the low-rank matrices.

Fields

lora_a: Tensor

Low-rank matrix A (input_dim x rank)

lora_b: Tensor

Low-rank matrix B (rank x output_dim)

scaling: f32

Scaling factor

active: bool

Whether this adapter is active

Implementations

impl LoRAAdapter

pub fn new( input_dim: usize, output_dim: usize, rank: usize, alpha: f32, ) -> Result<Self>

Creates a new LoRA adapter with random initialization.

pub fn forward(&self, input: &Tensor) -> Result<Tensor>

Compute the LoRA update: scaling * B @ A

pub fn get_delta_weight(&self) -> Result<Tensor>

Get the effective weight matrix ∆W = scaling * B @ A

pub fn merge_into_weight(&self, base_weight: &mut Tensor) -> Result<()>

Merge adapter weights into the base weight matrix

pub fn set_active(&mut self, active: bool)

Enable or disable this adapter

pub fn num_parameters(&self) -> usize

Get the number of trainable parameters

Trait Implementations

impl Clone for LoRAAdapter

fn clone(&self) -> LoRAAdapter

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LoRAAdapter

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

Formats the value using the given formatter.