Struct CompressedData 

pub struct CompressedData<T: Float + Debug + ScalarOperand + Send + Sync + 'static> {
    pub shape: Vec<usize>,
    pub codes: Vec<u8>,
    pub quantization_min: f64,
    pub quantization_step: f64,
    pub original_size: usize,
    pub compressed_size: usize,
    /* private fields */
}

A quantized tensor.

The payload is codes (one u8 per element) plus the affine dequantization parameters. The old version of this struct stored data: Vec<T> — the uncompressed input — while advertising a halved compressed_size.

Fields

shape: Vec<usize>

Original tensor shape (2-D).

codes: Vec<u8>

Quantization codes, row-major.

quantization_min: f64

Value the zero code maps to.

quantization_step: f64

Value increment per code step (0 for a constant tensor).

original_size: usize

Bytes the uncompressed tensor occupied.

compressed_size: usize

Bytes this payload occupies, counted for real.

Implementations

impl<T: Float + Debug + ScalarOperand + Send + Sync + 'static> CompressedData<T>

pub fn compression_ratio(&self) -> f64

Achieved compression ratio compressed_size / original_size.

A measurement, not the configured target: for f64 input this is roughly 1/8 plus metadata, and for f32 roughly 1/4.

pub fn max_reconstruction_error(&self) -> f64

Worst-case absolute reconstruction error, quantization_step / 2.

pub fn element_count(&self) -> usize

Number of quantized elements.

Trait Implementations

impl<T: Clone + Float + Debug + ScalarOperand + Send + Sync + 'static> Clone for CompressedData<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Debug + Float + Debug + ScalarOperand + Send + Sync + 'static> Debug for CompressedData<T>

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

Formats the value using the given formatter.