Enum QuantWeight 

pub enum QuantWeight {
    Nvfp4(QuantizedWeight),
    Fp8(Fp8Weight),
    Dense(DenseWeight),
    PackedQ2(PackedQ2Weight),
}

Quantized weight for any supported format.

Encapsulates all data a GEMV/GEMM kernel needs to dequantize and compute. The forward path matches on this enum to select the correct kernel. Enum branch compiles to ~1 cycle vs GPU kernel launch at ~5000 cycles.

Variants

Nvfp4(QuantizedWeight)

NVFP4 E2M1: packed nibbles + FP8 group scales + f32 global scale. Kernel: w4a16_gemv (decode) / w4a16_gemm (prefill)

Fp8(Fp8Weight)

FP8 E4M3: byte-packed weights + per-block BF16 scales. Kernel: w8a16_gemv (decode) / w8a16_gemm (prefill)

Dense(DenseWeight)

BF16 dense (unquantized). Kernel: dense_gemv / dense_gemm

PackedQ2(PackedQ2Weight)

Keep-packed ternary Q2_0 (ATLAS_GGUF_NATIVE_Q2): raw block_q2_0 bytes, 2-bit resident. Decode dispatches q2_0_gemv_vec; prefill transient- dequants to BF16 then runs dense_gemm. Tier-1c attention path.

Implementations

impl QuantWeight

pub fn null() -> Self

Null weight (for remote experts under EP or unused projections).

pub fn is_null(&self) -> bool

Whether this weight points to NULL (placeholder).

pub fn as_packed_q2(&self) -> Option<&PackedQ2Weight>

Extract as keep-packed Q2_0, if this weight is that variant.

pub fn as_nvfp4(&self) -> Option<&QuantizedWeight>

Extract as NVFP4, if this weight is that variant.

pub fn as_fp8(&self) -> Option<&Fp8Weight>

Extract as FP8, if this weight is that variant.

pub fn as_dense(&self) -> Option<&DenseWeight>

Extract as Dense, if this weight is that variant.

Trait Implementations

impl Clone for QuantWeight

fn clone(&self) -> QuantWeight

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantWeight

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

Formats the value using the given formatter.

impl From<DenseWeight> for QuantWeight

fn from(w: DenseWeight) -> Self

Converts to this type from the input type.

impl From<Fp8Weight> for QuantWeight

fn from(w: Fp8Weight) -> Self

Converts to this type from the input type.

impl From<PackedQ2Weight> for QuantWeight

fn from(w: PackedQ2Weight) -> Self

Converts to this type from the input type.

impl From<QuantizedWeight> for QuantWeight

fn from(w: QuantizedWeight) -> Self

Converts to this type from the input type.

impl Copy for QuantWeight