Struct TensorView 

pub struct TensorView<'a> { /* private fields */ }

A borrowed window onto one tensor, carrying its shape.

Implementations

impl<'a> TensorView<'a>

pub const fn dtype(&self) -> Dtype

Storage dtype.

pub fn shape(&self) -> &[usize]

Logical shape.

pub fn len(&self) -> usize

Element count.

pub fn is_empty(&self) -> bool

Whether the tensor has no elements.

pub fn row_len(&self) -> usize

Elements per row (product of all dimensions after the first).

pub fn get_f32(&self, index: usize) -> Option<f32>

Read one element, widened to f32.

Returns None if index is past the end. This is the accessor-level widening the design calls for: no whole-tensor f32 buffer is ever built.

pub fn copy_row_f32(&self, row: usize, out: &mut [f32]) -> bool

Copy one row, widening to f32, into out.

This is the cold-row path for the 151 936 × 2048 text embedding: prefill touches only the rows its token ids name, so only those rows are ever fetched and widened. Nothing advises the whole section resident.

Returns false (writing nothing) if the row is out of range or out is the wrong length, so a caller cannot silently consume a partially-filled buffer.

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

Borrow the raw, un-widened bytes.

pub const fn as_ref(&self) -> TensorViewRef<'a>

A shape-less reference to the same bytes.

Trait Implementations

impl<'a> Clone for TensorView<'a>

fn clone(&self) -> TensorView<'a>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<'a> Debug for TensorView<'a>

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

Formats the value using the given formatter.