Struct DsparkState 

pub struct DsparkState {
    pub rings: Vec<CudaSlice<f32>>,
    pub taps: CudaSlice<f32>,
}

DSpark decode-side state: the 3 per-block main_kv rings, each allocated [win + block_size, hd] — rows [0, win) are the persistent ring (slot = pos % win, M:783), rows [win, win+block) hold the CURRENT round’s transient draft kv (the M:784 cat([kv_cache, draft_kv]) gather realized in one buffer; rewritten every propose, never read as ring). Rings advance ONLY for committed positions (dspark_write_rings) — the §3.1 drafter rule.

Fields

rings: Vec<CudaSlice<f32>>

taps: CudaSlice<f32>

tap rows [t_max, n_targets*hidden] on the last stage: the hc-mean concat of layers 40/41/42, written by the decode step (and consumed by write_rings / forward_spec).