Struct DraftBinding 

pub struct DraftBinding {
    pub layer: CircuitLayer,
    pub k_pool: DevicePtr,
    pub v_pool: DevicePtr,
    pub block_size: u32,
    pub cache_stride: u64,
    pub vocab: u32,
    pub rows: Option<DraftRows>,
}

2026-09-29: The MTP draft head as the executor binds it: its one layer (weights by slot, the attention facts of its own KV cache) and that cache.

Fields

layer: CircuitLayer

2026-09-29: Weights by slot, the draft attention’s facts (attn_layer_idx is its cache’s layer), and what the circuit does not model.

k_pool: DevicePtr

2026-09-29: The draft cache’s K pool.

v_pool: DevicePtr

2026-09-29: The draft cache’s V pool.

block_size: u32

2026-09-29: Tokens per block of the draft cache.

cache_stride: u64

2026-09-29: PagedKvCache::cache_stride of the draft cache.

vocab: u32

2026-09-29: The vocabulary rows the draft lm_head scores and the argmax reads (the first mtp_vocab_size, or all).

rows: Option<DraftRows>

2026-09-30: What the head’s n-row draft (forward_batch_position) reads beyond the layer; None for a head outside the batched propose’s scope.

Trait Implementations

impl Clone for DraftBinding

fn clone(&self) -> DraftBinding

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for DraftBinding

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.