pub struct Glm5NextMtpModule {
pub layer: Glm5NextLayer,
pub eh_proj: DenseWeight,
pub enorm: DevicePtr,
pub hnorm: DevicePtr,
pub final_norm: DevicePtr,
}Expand description
2026-09-25: The MTP block plus the four tensors around it.
Fields§
§layer: Glm5NextLayer2026-09-25: The block as a mhc: None layer: DSA mixer, routed MoE, plain residual.
eh_proj: DenseWeight2026-09-25: [hidden, 2 * hidden]: projects concat(enorm(embed), hnorm(hidden))
back to hidden (glm5next_mtp_head.rs).
enorm: DevicePtr2026-09-25: RMSNorm weights for the embedding half and the target-hidden half of the concat.
hnorm: DevicePtr§final_norm: DevicePtr2026-09-25: shared_head.norm, the final norm before the target’s lm_head.
Auto Trait Implementations§
impl Freeze for Glm5NextMtpModule
impl RefUnwindSafe for Glm5NextMtpModule
impl Send for Glm5NextMtpModule
impl Sync for Glm5NextMtpModule
impl Unpin for Glm5NextMtpModule
impl UnwindSafe for Glm5NextMtpModule
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