Struct TensorSelector 

pub struct TensorSelector { /* private fields */ }

Exact bounded graph-node contract.

Implementations

impl TensorSelector

pub fn new( name: impl Into<String>, element_type: TensorElementType, row_elements: usize, maximum_rows: usize, access: TensorAccess, row_mapping: TensorRowMapping, retain: bool, ) -> Result<Self, TensorTransactionError>

Constructs an exact tensor selector.

Errors

Returns an error for an empty, NUL-containing, or excessive name, unusable dimensions, an excessive element bound, or mutable non-f32 data.

pub const fn with_finite_validation( self, finite: TensorFiniteValidation, ) -> Self

Sets how aggressively f32 values are validated as finite.

Defaults to TensorFiniteValidation::Strict. Relaxing this trades a full pass (or two, for ReadWriteF32) over the tensor for the caller’s guarantee that the values are already finite.

pub const fn finite_validation(&self) -> TensorFiniteValidation

Returns the finiteness-validation policy.

pub fn layer_output( layer: u32, row_elements: usize, maximum_rows: usize, access: TensorAccess, retain: bool, ) -> Result<Self, TensorTransactionError>

Constructs one exact residual layer-output selector.

llama.cpp names these graph nodes l_out-{layer} at the pinned implementation. The caller remains responsible for binding that naming profile to its backend revision.

Errors

Returns an error for unusable dimensions or bounds.

pub fn name(&self) -> &str

Returns the exact graph-node name.

pub const fn element_type(&self) -> TensorElementType

Returns the required element representation.

pub const fn row_elements(&self) -> usize

Returns the required elements per row.

pub const fn maximum_rows(&self) -> usize

Returns the inclusive row bound.

pub const fn access(&self) -> TensorAccess

Returns native access granted to the handler.

pub const fn row_mapping(&self) -> TensorRowMapping

Returns how native rows map to the submitted decode batch.

pub const fn retains_capture(&self) -> bool

Returns whether the completed owned tensor is retained after decode.

Trait Implementations

impl Clone for TensorSelector

fn clone(&self) -> TensorSelector

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorSelector

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

Formats the value using the given formatter.

impl Eq for TensorSelector

impl PartialEq for TensorSelector

fn eq(&self, other: &TensorSelector) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for TensorSelector