Struct BoftLayer 

pub struct BoftLayer { /* private fields */ }

BOFT layer implementing Butterfly Orthogonal Fine-Tuning.

Uses butterfly factorization of block-diagonal orthogonal matrices: W' = W @ R where R is constructed from butterfly factors.

Each butterfly factor is: P @ BlockDiag(O_i) @ P^T where P is a permutation matrix and O_i are orthogonal matrices.

Implementations

impl BoftLayer

pub fn new( in_features: usize, out_features: usize, config: BoftConfig, device: &Device, ) -> Result<BoftLayer, PeftError>

Create a new BOFT layer.

Arguments

• in_features - Input dimension
• out_features - Output dimension
• config - BOFT configuration
• device - Device to create tensors on

Errors

Returns error if configuration is invalid or tensor initialization fails.

pub fn block_num(&self) -> usize

Get the number of blocks.

pub fn block_size(&self) -> usize

Get the block size.

pub fn n_butterfly_factor(&self) -> usize

Get the number of butterfly factors.

Trait Implementations

impl Adapter for BoftLayer

type Config = BoftConfig

The configuration type for this adapter.

fn forward( &self, input: &Tensor, base_output: Option<&Tensor>, ) -> Result<Tensor, PeftError>

Forward pass applying the adapter transformation.

fn num_parameters(&self) -> usize

Get the number of trainable parameters.

fn config(&self) -> &<BoftLayer as Adapter>::Config

Get the adapter’s configuration.

impl Mergeable for BoftLayer

fn merge(&self, base_weight: &Tensor) -> Result<Tensor, PeftError>

Merge adapter weights into base model weights.

fn unmerge(&self, merged_weight: &Tensor) -> Result<Tensor, PeftError>

Unmerge adapter weights from merged weights.

impl SaveLoad for BoftLayer

fn state_dict(&self) -> Result<HashMap<String, Tensor>, PeftError>

Get all adapter tensors as a map of name -> tensor.

fn load_state_dict( &mut self, state_dict: HashMap<String, Tensor>, ) -> Result<(), PeftError>

Load adapter tensors from a state dict.

impl Trainable for BoftLayer

fn register_parameters( &self, var_map: &mut VarMap, prefix: &str, ) -> Result<(), PeftError>

Register trainable parameters with the variable map.

fn freeze(&mut self)

Freeze all adapter parameters (disable gradients).

fn unfreeze(&mut self)

Unfreeze all adapter parameters (enable gradients).

fn is_frozen(&self) -> bool

Check if the adapter is frozen.