Struct Lfm2Block 

pub struct Lfm2Block {
    pub operator_norm: RMSNorm,
    pub op: Lfm2LayerOp,
    pub ffn_norm: RMSNorm,
    pub mlp: MLP,
}

LFM2 block: x + op(norm(x)) then h + mlp(norm(h)).

Fields

operator_norm: RMSNorm

Norm before the operator.

op: Lfm2LayerOp

Convolution or attention.

ffn_norm: RMSNorm

Norm before the MLP.

mlp: MLP

Feed-forward.

Implementations

impl Lfm2Block

pub fn forward( &self, engine: &ComputeEngine, x: &Tensor, pos: usize, cache: Option<&KVCacheLayer>, mask: Option<&UnifiedTensor>, ) -> Tensor

Runs the block; pos, cache and mask are used by attention layers.

pub fn params(&self) -> Vec<Tensor>

Trainable tensors (weights, then any LoRA adapters).

pub fn lora_params(&self) -> Vec<Tensor>

Only the LoRA adapter tensors (what LoRA fine-tuning updates).

pub fn named_params(&self, prefix: &str) -> HashMap<String, Tensor>

Parameters keyed by Hugging Face-style names under prefix.

pub fn named_lora_params(&self, prefix: &str) -> HashMap<String, Tensor>

LoRA tensors keyed by name under prefix.