Struct Ia3Layer 

pub struct Ia3Layer { /* private fields */ }

IA³ layer implementing learned rescaling vectors.

For non-feedforward modules: output = base_output * ia3_vector For feedforward modules: output = base_layer(input * ia3_vector)

Implementations

impl Ia3Layer

pub fn new( in_features: usize, out_features: usize, is_feedforward: bool, config: Ia3Config, device: &Device, ) -> Result<Ia3Layer, PeftError>

Create a new IA³ layer.

Arguments

• in_features - Input dimension
• out_features - Output dimension
• is_feedforward - Whether this is a feedforward layer (scales input vs output)
• config - IA³ configuration
• device - Device to create tensors on

Errors

Returns error if configuration is invalid or tensor initialization fails.

pub fn scaling_vector(&self) -> &Tensor

Get the scaling vector.

pub fn is_feedforward(&self) -> bool

Check if this is a feedforward layer.

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

Apply IA³ scaling to input (for feedforward layers).

Arguments

• input - Input tensor [batch, seq_len, in_features]

Returns

Scaled input tensor

Errors

Returns error if called on non-feedforward layer or tensor operations fail.

pub fn scale_output(&self, output: &Tensor) -> Result<Tensor, PeftError>

Apply IA³ scaling to output (for non-feedforward layers).

Arguments

• output - Output tensor from base layer [batch, seq_len, out_features]

Returns

Scaled output tensor

Errors

Returns error if called on feedforward layer or tensor operations fail.

Trait Implementations

impl Adapter for Ia3Layer

type Config = Ia3Config

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) -> &<Ia3Layer as Adapter>::Config

Get the adapter’s configuration.

impl Mergeable for Ia3Layer

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 Trainable for Ia3Layer

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.