Struct DequantizePerGroupPlan 

pub struct DequantizePerGroupPlan<TIn: Element, TOut: IntElement> { /* private fields */ }

dequantize_per_group plan.

x[..., g] = scale[outer, g] * (q[..., g] - zero_point[outer, g]) with g the group index along the (rightmost) quant axis. Inverse of QuantizePerGroupPlan.

When to use: FP recovery from INT4/INT8 grouped weight blobs (GPTQ / AWQ / GGML). Pair with DequantizePerGroupBackwardPlan.

Dtypes: input int {s8, u8}; output FP {f32, f64, f16, bf16}.

Shape limits: rank-2 [outer, axis_size] with axis_size % group_size == 0.

Workspace: none.

Precision guarantee: deterministic, bit-stable.

Implementations

impl<TIn: Element, TOut: IntElement> DequantizePerGroupPlan<TIn, TOut>

pub fn select( _stream: &Stream, desc: &DequantizePerGroupDescriptor, _pref: PlanPreference, ) -> Result<Self>

Pick a kernel for desc.

pub fn can_implement( &self, args: &DequantizePerGroupArgs<'_, TIn, TOut>, ) -> Result<()>

Validate args.

pub fn workspace_size(&self) -> usize

Workspace bytes — none.

pub fn sku(&self) -> KernelSku

Identity.

pub fn precision_guarantee(&self) -> PrecisionGuarantee

Numerical guarantees.

pub fn run( &self, stream: &Stream, _workspace: Workspace<'_>, args: DequantizePerGroupArgs<'_, TIn, TOut>, ) -> Result<()>

Launch.