Struct DflashConfig 

pub struct DflashConfig {
    pub draft_model: &'static str,
    pub gamma: usize,
    pub window_size: usize,
    pub mask_token_id: u32,
    pub target_layer_ids: &'static [usize],
}

DFlash speculative-decoding pairing for a target model. Parsed from [dflash] in MODEL.toml at build time. None when the model has no DFlash drafter associated.

Fields

draft_model: &'static str

HuggingFace id (or local path) of the drafter checkpoint.

gamma: usize

Block size γ (parallel draft tokens per step). Defaults to 16.

window_size: usize

Drafter sliding-window size in tokens. 0 = full attention.

mask_token_id: u32

Token id used to fill the γ “to-be-predicted” positions during drafter forward. From the drafter’s dflash_config.mask_token_id.

target_layer_ids: &'static [usize]

Target-side layer indices to capture intermediate hidden states from (shallow-to-deep). The drafter’s fc projection consumes the stack of these hiddens. From the drafter’s dflash_config.target_layer_ids.

Trait Implementations

impl Clone for DflashConfig

fn clone(&self) -> DflashConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DflashConfig

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

Formats the value using the given formatter.