pub struct ProposeGraphs {
pub by_width: HashMap<usize, Vec<GraphHandle>>,
pub warmup: HashMap<usize, usize>,
}Expand description
Block-diffusion draft head. Public API is the DraftProposer trait.
The drafter shares embed_tokens and lm_head with the target — these
are NOT in the drafter’s safetensors checkpoint (verified against
z-lab/Qwen3.6-35B-A3B-DFlash commit 42d3b34). The constructor takes
the target’s embed_tokens_shared and lm_head_shared device pointers
at build time and slots them in alongside the drafter’s own fc,
hidden_norm, norm, and per-layer weights.
Per-block-width propose graph state (see BlockDiffusionDraftHead::propose_graphs).
warmup counts the eager passes run at a width before its capture
(default target 2, AVAROK_DFLASH_PROPOSE_WARMUP_N): two eager passes warm
the PTX->SASS cache, ramp GB10 clocks and bring hot weight tiles into L2
before the capture freezes the SASS variants the driver picks.
Fields§
§by_width: HashMap<usize, Vec<GraphHandle>>§warmup: HashMap<usize, usize>