pub struct MlaGeom {
pub n_head: usize,
pub d_nope: usize,
pub d_rope: usize,
pub d_v: usize,
pub kv_rank: usize,
pub latent_dim: usize,
pub scale: f32,
}Expand description
Latent-KV geometry for one MLA layer, resolved at load from MlaConfig (glm-dsa). The KV
cache stores ONE latent_dim-wide row per token per layer: [rmsnorm(c_kv) | rope(k_pe)];
V is the first kv_rank elements of the SAME row (no V plane). All heads stream it (MQA).
Fields§
§n_head: usize§d_nope: usize§d_rope: usize§d_v: usize§kv_rank: usize§latent_dim: usize§scale: f32Trait Implementations§
Auto Trait Implementations§
impl Freeze for MlaGeom
impl RefUnwindSafe for MlaGeom
impl Send for MlaGeom
impl Sync for MlaGeom
impl Unpin for MlaGeom
impl UnsafeUnpin for MlaGeom
impl UnwindSafe for MlaGeom
Blanket Implementations§
Source§impl<T> BorrowMut<T> for Twhere
T: ?Sized,
impl<T> BorrowMut<T> for Twhere
T: ?Sized,
Source§fn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
Mutably borrows from an owned value. Read more