Struct TensorViewWithDataBuffer

pub struct TensorViewWithDataBuffer { /* private fields */ }

A data structure like TensorView in safetensors, but it owns the data

Implementations

impl TensorViewWithDataBuffer

pub fn to_tensor_view(&self) -> TensorView<'_>

Create a standard TensorView from this buffer. No copy of data occurs.

pub fn new<A, S, D>(array: &ArrayBase<S, D>) -> TensorViewWithDataBuffer

where A: CommonSupportedElement, S: Data<Elem = A>, D: Dimension,

Create a new TensorViewWithDataBuffer object from a ndarray.

Notes: Copy of the data occurs once. The original array won’t be consumed.

pub fn new_fp16<A, S, D>(array: &ArrayBase<S, D>) -> TensorViewWithDataBuffer

where A: Float16ConversionSupportedElement, S: Data<Elem = A>, D: Dimension,

Create a new TensorViewWithDataBuffer object in FP16 from a ndarray.

pub fn new_bf16<A, S, D>(array: &ArrayBase<S, D>) -> TensorViewWithDataBuffer

where A: BFloat16ConversionSupportedElement, S: Data<Elem = A>, D: Dimension,

Create a new TensorViewWithDataBuffer object in BF16 from a ndarray.

Trait Implementations

impl View for TensorViewWithDataBuffer

fn dtype(&self) -> Dtype

The Dtype of the tensor

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

The shape of the tensor

fn data(&self) -> Cow<'_, [u8]>

The data of the tensor

fn data_len(&self) -> usize

The length of the data, in bytes. This is necessary as this might be faster to get than data().len() for instance for tensors residing in GPU.