Struct QuantizedTensorGeneric 

pub struct QuantizedTensorGeneric<R: QuantRange> { /* private fields */ }

Generic quantized tensor, parameterized by bit-width marker.

For INT4 tensors, call QuantizedTensorGeneric::pack to compress two values per byte for 2× storage savings.

Implementations

impl<R: QuantRange> QuantizedTensorGeneric<R>

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

Tensor shape.

pub fn params(&self) -> &QuantParamsGeneric<R>

Per-tensor quantization parameters (channel-0 if per-channel).

pub fn is_per_channel(&self) -> bool

Whether per-channel quantization was used.

pub fn from_f32(data: &[f32], shape: Vec<usize>) -> Result<Self>

Quantize FP32 data, computing the range from the data.

Errors

Returns QuantizeError::InvalidTensor if data is empty or shape mismatches.

pub fn from_f32_with_range( data: &[f32], shape: Vec<usize>, min: f32, max: f32, ) -> Result<Self>

Quantize FP32 data using an explicit range (for calibration).

Errors

Returns QuantizeError::InvalidTensor if data is empty or shape mismatches.

pub fn from_f32_per_channel(data: &[f32], shape: Vec<usize>) -> Result<Self>

Quantize FP32 data with per-channel ranges (axis 0 only).

Errors

Returns QuantizeError::InvalidTensor if data is empty, shape mismatches, or the tensor is scalar.

pub fn to_f32(&self) -> Vec<f32>

Dequantize all values back to FP32.

pub fn size_bytes(&self) -> usize

Size of the quantized data in bytes (packed if available, unpacked otherwise).

pub fn quantization_error(&self, original: &[f32]) -> f32

Mean squared error between the original data and the dequantized values.

impl QuantizedTensorGeneric<Int4Range>

pub fn pack(&mut self)

Pack two INT4 values per byte for 2× compression.

pub fn ensure_unpacked(&self) -> Vec<i8>

Return unpacked i8 data, decompressing from packed storage if needed.

pub fn is_packed(&self) -> bool

Whether the data is currently bit-packed.

pub fn packed_size_bytes(&self) -> usize

Size that the packed representation would occupy (or already occupies).

pub fn unpacked_size_bytes(&self) -> usize

Size of the unpacked representation in bytes.

Trait Implementations

impl<R: Clone + QuantRange> Clone for QuantizedTensorGeneric<R>

fn clone(&self) -> QuantizedTensorGeneric<R>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<R: Debug + QuantRange> Debug for QuantizedTensorGeneric<R>

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

Formats the value using the given formatter.