Struct TurboQuantizedView 

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

Available on crate feature alloc only.

Zero-copy view over a TurboQuant packed binary.

Constructed from &[u8] with no allocation – suitable for embedded targets where the binary is in flash/ROM. Note: predict does allocate for the Hadamard rotation of the feature vector; use predict_with_scratch for zero-alloc inference.

Implementations

impl<'a> TurboQuantizedView<'a>

pub fn from_bytes(bytes: &'a [u8]) -> Result<Self, FormatError>

Parse a TurboQuant binary from raw bytes.

Returns FormatError::Truncated if the buffer is too short, has inconsistent length, or contains an unknown quantization mode.

pub fn predict(&self, features: &[f64]) -> f64

Dot product of quantized weights with a feature vector.

Applies the same Hadamard rotation to features, then computes the dot product with the quantized rotated weights over the full padded length.

pub fn predict_with_scratch(&self, features: &[f64], scratch: &mut [f64]) -> f64

Predict using a caller-provided scratch buffer for the Hadamard rotation.

scratch must have length >= padded_len. This avoids allocation, making it suitable for embedded inference loops.

pub fn n_weights(&self) -> usize

Number of weights in this view.

pub fn padded_len(&self) -> usize

Power-of-2 padded length used for FWHT.

pub fn mode(&self) -> QuantMode

Quantization mode.