Struct KimiStoredExpertLayout 

pub struct KimiStoredExpertLayout {
    pub moe_hidden_dim: usize,
    pub moe_intermediate_dim: usize,
}

Loads one routed expert’s real MXFP4 weights ({prefix}.experts.{expert_idx}.{w1,w2,w3}.{weight_packed,weight_scale}). moe_hidden_dim is the real latent dimension (routed_expert_hidden_size=3584 for Kimi K3, not the outer hidden_dim=7168 – see ferrox-models::latent_moe’s module doc comment); moe_intermediate_dim is the per-expert FFN size (3072). Per-layer byte layout of one store-backed Kimi routed expert’s combined buffer: w1_packed, w1_scale, w2_packed, w2_scale, w3_packed, w3_scale concatenated in that fixed order. Every expert in a Kimi layer has identical dims, so one layout serves the layer.

Fields

moe_hidden_dim: usize

moe_intermediate_dim: usize

Implementations

impl KimiStoredExpertLayout

pub fn total_bytes(&self) -> usize

pub fn materialize(&self, lease: &ExpertLease) -> KimiExpertWeights

Builds temporary two-buffer MXFP4 WeightMatrix views over a leased buffer; each view’s WeightBytes::Shared clone keeps the cache entry pinned for the view’s lifetime.

Trait Implementations

impl Clone for KimiStoredExpertLayout

fn clone(&self) -> KimiStoredExpertLayout

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for KimiStoredExpertLayout

impl Debug for KimiStoredExpertLayout

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

Formats the value using the given formatter.