Struct FullAttentionLayer 

pub struct FullAttentionLayer<'a, Q: QuantWeights> {
    pub input_ln: DevicePtr,
    pub q_norm: DevicePtr,
    pub k_norm: DevicePtr,
    pub post_ln: DevicePtr,
    pub q_proj: &'a Q,
    pub k_proj: &'a Q,
    pub v_proj: &'a Q,
    pub o_proj: &'a Q,
    pub gate_proj: &'a Q,
    pub up_proj: &'a Q,
    pub down_proj: &'a Q,
}

Full-attention layer weights, parameterised over the backend’s quantised weight type.

Fields

input_ln: DevicePtr

q_norm: DevicePtr

k_norm: DevicePtr

post_ln: DevicePtr

q_proj: &'a Q

k_proj: &'a Q

v_proj: &'a Q

o_proj: &'a Q

gate_proj: &'a Q

up_proj: &'a Q

down_proj: &'a Q