Struct BlockDef 

pub struct BlockDef {
    pub name: String,
    pub attention: AttentionDef,
    pub ssm: Option<SsmDef>,
    pub ffn: FfnDef,
    pub norm: NormConfig,
    pub norm_position: NormPosition,
    pub residual: bool,
    pub dropout: f64,
}

Transformer block definition

The mixer is attention by default; when ssm is set the block is a Mamba block (attention settings are ignored).

Fields

name: String

attention: AttentionDef

ssm: Option<SsmDef>

ffn: FfnDef

norm: NormConfig

norm_position: NormPosition

residual: bool

dropout: f64

Implementations

impl BlockDef

pub fn is_ssm(&self) -> bool

pub fn num_heads(&self) -> usize

pub fn num_kv_heads(&self) -> usize

pub fn head_dim(&self, hidden_size: usize) -> usize

pub fn intermediate_size(&self, hidden_size: usize) -> usize

pub fn norm_eps(&self) -> f64

pub fn rope_theta(&self) -> f64

pub fn rope_scaling(&self) -> Option<f64>

Trait Implementations

impl Clone for BlockDef

fn clone(&self) -> BlockDef

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BlockDef

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

Formats the value using the given formatter.

impl Default for BlockDef

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for BlockDef

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for BlockDef

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.