Enum Precision 

pub enum Precision {
    F32,
    F16LmHead,
}

Precision policy for the Qwen3 inference graph. Today only F32 is exact; the others toggle the corresponding env-vars on the Metal MPSGraph fast path (see qwen3_metal_perf notes).

Variants

F32

Everything in F32. Default — most reproducible, slowest on large LM heads.

F16LmHead

F32 throughout except the LM-head matmul, which casts to F16 for the dominant prefill workload. Wins ~1.3-1.45× on (B≥2, L≥64) cells; loses on small cells.

Trait Implementations

impl Clone for Precision

fn clone(&self) -> Precision

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Precision

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

Formats the value using the given formatter.

impl Default for Precision

fn default() -> Precision

Returns the “default value” for a type.

impl PartialEq for Precision

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

impl Eq for Precision

impl StructuralPartialEq for Precision