Enum ResidualsConfig 

pub enum ResidualsConfig {
    Standard,
    MultiGate {
        n_stream: usize,
        init_bias: f64,
        per_virtual_layer: bool,
    },
}

Configuration / factory for Residuals.

Variants

Standard

Plain additive Pre-LN residual.

MultiGate

Multi-Gate Residuals over n_stream streams.

Fields

n_stream: usize

Number of parallel residual streams n.

init_bias: f64

Initial gate bias (see MultiGateResidualConfig::init_bias).

per_virtual_layer: bool

true ⇒ one MGR per virtual layer; false ⇒ one per real layer (reused across virtual passes). See MultiGate::per_virtual.

Implementations

impl ResidualsConfig

pub fn init( &self, d_model: usize, n_real_layers: usize, n_virtual_layers: usize, device: &Device, ) -> Residuals

Build the runtime Residuals for a stack of n_real_layers real weight sets unrolled over n_virtual_layers (virtual) passes. The MGR module count follows per_virtual_layer (one per virtual layer vs one per real layer).

Trait Implementations

impl Clone for ResidualsConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Config for ResidualsConfig

fn load_binary(data: &[u8]) -> Result<Self, ConfigError>

Loads the configuration from a binary buffer.

impl Debug for ResidualsConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ResidualsConfig

fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for ResidualsConfig

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

Formats the value using the given formatter.

impl Serialize for ResidualsConfig

fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>

where S: Serializer,

Serialize this value into the given Serde serializer.