Struct LinearAttnConfig 

pub struct LinearAttnConfig {
    pub layer_types: Vec<String>,
    pub conv_kernel_dim: usize,
    pub num_key_heads: usize,
    pub key_head_dim: usize,
    pub num_value_heads: usize,
    pub value_head_dim: usize,
}

Configuration for linear (recurrent) attention layers (e.g. Gated DeltaNet in Qwen3.5).

Fields

layer_types: Vec<String>

Per-layer type: “linear_attention” or “full_attention”.

conv_kernel_dim: usize

Conv1d kernel size for short convolution preprocessing.

num_key_heads: usize

Number of key heads in linear attention.

key_head_dim: usize

Per-head key dimension in linear attention.

num_value_heads: usize

Number of value heads in linear attention.

value_head_dim: usize

Per-head value dimension in linear attention.

Trait Implementations

impl Clone for LinearAttnConfig

fn clone(&self) -> LinearAttnConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LinearAttnConfig

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

Formats the value using the given formatter.