Struct AttentionWeights 

pub struct AttentionWeights {
    pub q_proj: DenseWeight,
    pub k_proj: DenseWeight,
    pub v_proj: DenseWeight,
    pub o_proj: QuantizedWeight,
    pub q_norm: DenseWeight,
    pub k_norm: DenseWeight,
    pub q_norm_full: Option<DenseWeight>,
    pub k_norm_full: Option<DenseWeight>,
    pub k_scale: f32,
    pub v_scale: f32,
}

2026-09-25: Full-attention layer weights.

Fields

q_proj: DenseWeight

k_proj: DenseWeight

v_proj: DenseWeight

o_proj: QuantizedWeight

q_norm: DenseWeight

2026-09-25: Per-head Q RMSNorm weight, [head_dim].

k_norm: DenseWeight

2026-09-25: Per-head K RMSNorm weight, [head_dim].

q_norm_full: Option<DenseWeight>

2026-09-25: Full-width Q RMSNorm weight, [num_heads * head_dim]: one norm over the whole Q projection before RoPE. The MiniMax loader sets it; when Some, the attention forward applies it instead of the per-head q_norm.

k_norm_full: Option<DenseWeight>

2026-09-25: Full-width K RMSNorm weight, [num_kv_heads * head_dim].

k_scale: f32

2026-09-25: K scale for the FP8 KV cache, used when no FP8 calibration is loaded.

v_scale: f32

2026-09-25: V scale for the FP8 KV cache, as k_scale.

Trait Implementations

impl Clone for AttentionWeights

fn clone(&self) -> AttentionWeights

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for AttentionWeights

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

Formats the value using the given formatter.

impl Copy for AttentionWeights