Struct WeightQuantTier 

pub struct WeightQuantTier { /* private fields */ }

2026-09-28: What the serve publishes: the tier and, under nvfp4, its W4A4 lever.

Implementations

impl WeightQuantTier

pub fn new(tier: WeightQuantization, downcast: W4a4Downcast) -> Result<Self>

2026-09-28: The nvfp4 tier with its W4A4 lever. A declared tier refuses a lever: the checkpoint already decides every layer’s activations there.

pub fn tier(self) -> WeightQuantization

2026-09-28: The tier.

pub fn downcast(self) -> W4a4Downcast

2026-09-28: The nvfp4 tier’s W4A4 lever; always Off under declared.

pub fn uses_w4a4_decode(self) -> bool

2026-09-28: Whether decode may take the FP4 block-scale projection path at all, so its kernels and activation scratch are prepared.

pub fn w4a4_rows( self, act: Nvfp4Act, narrow_rows: u32, wide_rows: u32, kernel_rows: u32, ) -> u32

2026-09-28: The widest row count the W4A4 path serves for a weight stamped act, given kernel_rows, the widest the prepared kernels serve (0 when absent). 0 means the weight’s decode activations stay 16-bit.

Under nvfp4 the lever decides for every weight alike: 32 rows, or 64 with -wide. Under declared a weight stamped Nvfp4Act::A4 runs W4A4 at every row count the kernels serve, and any other weight runs 16-bit activations.

pub fn ffn_small_batch_w4a4(self, any_a4: bool) -> bool

2026-09-28: Whether a dense FFN whose gate, up or down is stamped act sends its 2- and 3-row decode steps through the W4A4 projections. Under declared the stamps decide; under nvfp4 the lever does, as --w4a4-downcast always has.

pub fn ffn_single_row_w4a4(self, any_a4: bool) -> bool

2026-09-28: Whether a dense FFN runs its single decode row W4A4. Only declared does, and only for an FFN that declares FP4 activations; --w4a4-downcast never reached the single row.

Trait Implementations

impl Clone for WeightQuantTier

fn clone(&self) -> WeightQuantTier

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for WeightQuantTier

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

Formats the value using the given formatter.

impl Default for WeightQuantTier

fn default() -> WeightQuantTier

Returns the “default value” for a type.

impl PartialEq for WeightQuantTier

fn eq(&self, other: &WeightQuantTier) -> 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 WeightQuantTier

impl Eq for WeightQuantTier

impl StructuralPartialEq for WeightQuantTier