Enum MambaVocabNetConfig 

pub enum MambaVocabNetConfig {
    Mamba1 {
        n_real_layers: usize,
        n_virtual_layers: Option<(usize, Schedule)>,
        vocab_size: usize,
        pad_vocab_size_multiple: usize,
        mamba_block: Mamba1Config,
        missing_lm_head: bool,
        ignore_first_residual: bool,
        ignore_last_residual: bool,
        residuals: ResidualsConfig,
    },
    Mamba2 {
        n_real_layers: usize,
        n_virtual_layers: Option<(usize, Schedule)>,
        vocab_size: usize,
        pad_vocab_size_multiple: usize,
        mamba_block: Mamba2Config,
        missing_lm_head: bool,
        ignore_first_residual: bool,
        ignore_last_residual: bool,
        residuals: ResidualsConfig,
    },
    Mamba3 {
        n_real_layers: usize,
        n_virtual_layers: Option<(usize, Schedule)>,
        vocab_size: usize,
        pad_vocab_size_multiple: usize,
        mamba_block: Mamba3Config,
        missing_lm_head: bool,
        ignore_first_residual: bool,
        ignore_last_residual: bool,
        residuals: ResidualsConfig,
    },
}

The serializable, documentation-friendly config for MambaVocabNet. Each variant is concrete (per-family), so #[derive(Config)] applies; init builds the matching network variant.

Variants

Mamba1

Build a Mamba-1 language model.

Fields

n_real_layers: usize

Number of real layers.

n_virtual_layers: Option<(usize, Schedule)>

Optional virtual-layer scheduling.

vocab_size: usize

Unpadded vocabulary size.

pad_vocab_size_multiple: usize

Round vocab_size up to a multiple of this (1 disables rounding).

mamba_block: Mamba1Config

Shared block config.

missing_lm_head: bool

Tie the LM head to the (transposed) embedding weights when true.

ignore_first_residual: bool

Suppress the first virtual layer’s residual (Pre-LN skip / MultiGate seed carry). See Layers.

ignore_last_residual: bool

Suppress the last virtual layer’s residual (output is the last layer’s transform alone). See Layers.

residuals: ResidualsConfig

Inter-layer residual scheme (plain additive vs Multi-Gate).

Mamba2

Build a Mamba-2 language model.

Fields

n_real_layers: usize

Number of real layers.

n_virtual_layers: Option<(usize, Schedule)>

Optional virtual-layer scheduling.

vocab_size: usize

Unpadded vocabulary size.

pad_vocab_size_multiple: usize

Round vocab_size up to a multiple of this (1 disables rounding).

mamba_block: Mamba2Config

Shared block config.

missing_lm_head: bool

Tie the LM head to the (transposed) embedding weights when true.

ignore_first_residual: bool

Suppress the first virtual layer’s residual (Pre-LN skip / MultiGate seed carry). See Layers.

ignore_last_residual: bool

Suppress the last virtual layer’s residual (output is the last layer’s transform alone). See Layers.

residuals: ResidualsConfig

Inter-layer residual scheme (plain additive vs Multi-Gate).

Mamba3

Build a Mamba-3 language model.

Fields

n_real_layers: usize

Number of real layers.

n_virtual_layers: Option<(usize, Schedule)>

Optional virtual-layer scheduling.

vocab_size: usize

Unpadded vocabulary size.

pad_vocab_size_multiple: usize

Round vocab_size up to a multiple of this (1 disables rounding).

mamba_block: Mamba3Config

Shared block config.

missing_lm_head: bool

Tie the LM head to the (transposed) embedding weights when true.

ignore_first_residual: bool

Suppress the first virtual layer’s residual (Pre-LN skip / MultiGate seed carry). See Layers.

ignore_last_residual: bool

Suppress the last virtual layer’s residual (output is the last layer’s transform alone). See Layers.

residuals: ResidualsConfig

Inter-layer residual scheme (plain additive vs Multi-Gate).

Implementations

impl MambaVocabNetConfig

pub fn init(&self, device: &Device) -> MambaVocabNet

Allocate and initialise the selected language model on device.

Trait Implementations

impl Clone for MambaVocabNetConfig

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 MambaVocabNetConfig

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

Loads the configuration from a binary buffer.

impl Debug for MambaVocabNetConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for MambaVocabNetConfig

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

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for MambaVocabNetConfig

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

Formats the value using the given formatter.

impl Serialize for MambaVocabNetConfig

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

where S: Serializer,

Serialize this value into the given Serde serializer.