Struct TensorData 

pub struct TensorData {
    pub name: Option<String>,
    pub dtype: DataType,
    pub dims: Vec<usize>,
    pub data: Vec<u8>,
    pub strings: Vec<String>,
}

A concrete constant tensor held inline (e.g. an attribute value or a small initializer). Element bytes are stored little-endian and densely packed.

Large model weights are referenced lazily via WeightRef instead.

Fields

name: Option<String>

dtype: DataType

dims: Vec<usize>

Static dimensions (constants always have a fully known shape).

data: Vec<u8>

Raw little-endian element bytes. Sub-byte values are densely packed (two 4-bit or four 2-bit elements per byte); for DataType::String this is empty and strings is used instead.

strings: Vec<String>

String payloads for DataType::String tensors.

Implementations

impl TensorData

pub fn from_raw(dtype: DataType, dims: Vec<usize>, data: Vec<u8>) -> Self

A numeric tensor from raw little-endian bytes.

pub fn numel(&self) -> usize

Number of elements (product of dims; 1 for a scalar).

pub fn expected_bytes(&self) -> usize

Expected byte length for numel elements of dtype, accounting for sub-byte packing.

Trait Implementations

impl Clone for TensorData

fn clone(&self) -> TensorData

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorData

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

Formats the value using the given formatter.

impl PartialEq for TensorData

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for TensorData