pub struct DraftRows {
pub meta: DevicePtr,
pub lp_offset: usize,
pub lm_head_gemv: Vec<KernelHandle>,
pub lm_head_twin: bool,
}Expand description
2026-09-30: The MTP head’s n-row draft facts: where its batched metadata and confidences
live, and the LM-head launch it chooses at each width (MtpHead::lm_head_batch_kernel,
mtp_head::lm_head_rows_arm), which the executor checks the plan against.
Fields§
§meta: DevicePtr2026-09-30: The batched propose’s metadata allocation (MtpHead::propose_meta).
lp_offset: usize2026-09-30: Byte offset in scratch of the per-row top-1 log-probabilities
(LP_SCRATCH_OFF); the ids sit at scratch’s start.
lm_head_gemv: Vec<KernelHandle>2026-09-30: lm_head_batch_kernel(n) at index n; the zero handle where none serves.
lm_head_twin: bool2026-09-30: The tile GEMM on the transposed LM-head twin is available to the head.
Trait Implementations§
Auto Trait Implementations§
impl Freeze for DraftRows
impl RefUnwindSafe for DraftRows
impl Send for DraftRows
impl Sync for DraftRows
impl Unpin for DraftRows
impl UnwindSafe for DraftRows
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