Enum DflashQuantization 

pub enum DflashQuantization {
    Bf16,
    Fp8Weights,
}

Drafter-side weight precision. Defaults to BF16. Phase G (2026-05-28) adds Fp8Weights, gated by env var ATLAS_DFLASH_DRAFTER_FP8. The historical SM12.x acceptance collapse note applied to drafter FP8 KV cache (different concern — bidirectional attention math); Phase G targets weight FP8 only, so the risk surface is dynamic-range loss in MLP intermediate activations, which per-row scales mitigate. --mtp-quantization fp8 is still not honored for the DFlash drafter.

Variants

Bf16

Fp8Weights

Weight-only FP8: q/k/v/o/gate/up/down BF16 → FP8 E4M3 with per-row f32 scales at model load. Activations stay BF16; KV cache stays BF16. GEMMs use fp8_gemm_n128 (BF16 × FP8 → BF16).

Trait Implementations

impl Clone for DflashQuantization

fn clone(&self) -> DflashQuantization

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DflashQuantization

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

Formats the value using the given formatter.

impl PartialEq for DflashQuantization

fn eq(&self, other: &DflashQuantization) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Copy for DflashQuantization

impl Eq for DflashQuantization

impl StructuralPartialEq for DflashQuantization