Struct Nvfp4Artifact 

pub struct Nvfp4Artifact { /* private fields */ }

A validated F32 × NVFP4 projection artifact.

Implementations

impl Nvfp4Artifact

pub fn new( m: u32, n: u32, k: u32, activation: Nvfp4Activation, ) -> Result<Self, LoweringError>

Describe [m,k] F32 × [n,k] NVFP4 -> [m,n] F32.

The tensor scale is a one-element F32 input rather than artifact metadata, so one loaded program serves every checkpoint tensor with this geometry.

pub fn new_batched( batch: u32, n: u32, k: u32, activation: Nvfp4Activation, ) -> Result<Self, LoweringError>

Describe batch independent F32 vectors of length k multiplied by NVFP4 weight matrices with n rows and k columns.

Activations, packed weights, block scales and tensor scales all carry the leading batch dimension. This is the natural execution unit for a routed MoE layer and amortizes submission over all selected experts.

pub fn new_expert_batch( batch: u32, n: u32, k: u32, activation: Nvfp4Activation, ) -> Result<Self, LoweringError>

Describe an expert batch whose packed weights and block scales are separate bindings. This retains zero-copy imports for independently cached experts while still issuing one dispatch. The Vulkan descriptor budget permits up to six experts.

pub fn as_bytes(&self) -> &[u8]

pub fn as_ref(&self) -> ArtifactRef<'_>

Trait Implementations

impl Clone for Nvfp4Artifact

fn clone(&self) -> Nvfp4Artifact

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for Nvfp4Artifact

impl Debug for Nvfp4Artifact

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

Formats the value using the given formatter.

impl PartialEq for Nvfp4Artifact

fn eq(&self, other: &Nvfp4Artifact) -> 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 StructuralPartialEq for Nvfp4Artifact