Struct Tensor 

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

Read-only borrow of a tensor owned by a crate::Interpreter.

Tensor<'a> cannot outlive the interpreter it came from — the borrow checker enforces this by attaching a phantom lifetime to the underlying pointer.

Implementations

impl<'a> Tensor<'a>

pub fn dims(&self) -> Vec<i32>

Tensor shape as a list of per-axis dimensions.

pub fn dtype(&self) -> TfLiteType

Element type of the tensor.

pub fn name(&self) -> Option<&str>

Tensor name as reported by the model, if any.

pub fn byte_size(&self) -> usize

Total size of the tensor’s data buffer in bytes.

pub fn quantization(&self) -> TfLiteQuantizationParams

Affine quantization parameters (scale, zero_point).

pub fn data(&self) -> &[u8]

Raw byte view of the tensor’s data buffer.

Returns an empty slice if the C API reports a NULL data pointer.

pub fn as_slice_f32(&self) -> Result<&[f32]>

Typed &[f32] view; errors if the tensor is not Float32.

pub fn as_slice_i8(&self) -> Result<&[i8]>

Typed &[i8] view; errors if the tensor is not Int8.

pub fn as_slice_u8(&self) -> Result<&[u8]>

Typed &[u8] view; errors if the tensor is not UInt8.

pub fn as_slice_i32(&self) -> Result<&[i32]>

Typed &[i32] view; errors if the tensor is not Int32.

pub fn to_vec_f32(&self) -> Result<Vec<f32>>

Read the tensor as a fresh Vec<f32>, dequantizing Int8/UInt8 using the tensor’s quantization parameters.

The dequantization formula is real = (q - zero_point) * scale. Float32 tensors are copied straight through.