#[repr(usize)]pub enum Slot {
Qkv = 0,
AttnCore = 1,
AttnO = 2,
FfnGateUp = 3,
FfnDown = 4,
Norms = 5,
Head = 6,
Sampler = 7,
Layers = 8,
Matmat = 9,
PrefillAttend = 10,
SplitAct = 11,
}Variants§
Qkv = 0
Q/K/V projections of one decode position.
AttnCore = 1
RoPE + append + attend (the attention core, one position).
AttnO = 2
Output projection.
FfnGateUp = 3
Fused gate/up (+SiLU·mul).
FfnDown = 4
Down projection.
Norms = 5
RMSNorms + residual adds of the layer loop.
Head = 6
Final norm + lm_head matvec.
Sampler = 7
Sampler (argmax / top-k / penalties).
Layers = 8
Whole layer stack of one forward (decode or prefill chunk).
Matmat = 9
Batched prefill projections (every matmat).
PrefillAttend = 10
Batched prefill attention core (per-position or batched attend).
SplitAct = 11
Activation quantization (split_act) inside the kernels.
Trait Implementations§
Auto Trait Implementations§
impl Freeze for Slot
impl RefUnwindSafe for Slot
impl Send for Slot
impl Sync for Slot
impl Unpin for Slot
impl UnsafeUnpin for Slot
impl UnwindSafe for Slot
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