fused_quantized_matvec_sequence

scirs2_linalg::quantization::fusion

Function fused_quantized_matvec_sequence 

Source 
pub fn fused_quantized_matvec_sequence<F>(
    matrices: &[&QuantizedMatrix],
    matrix_params: &[&QuantizationParams],
    vector: &ArrayView1<'_, F>,
    output_quantize: bool,
) -> LinalgResult<Array1<F>>
where
    F: Float + Debug + AsPrimitive<f32> + FromPrimitive,
    f32: AsPrimitive<F>,
Expand description

Fused quantized matrix-vector multiplication sequence

Computes the matrix-vector sequence (A * B * … * x) where matrices and vector are in quantized form. This avoids dequantizing and requantizing intermediate results.

§Arguments

  • matrices - A slice of quantized matrices to multiply
  • matrix_params - A slice of quantization parameters for each matrix
  • vector - The quantized vector to multiply with
  • vector_params - Quantization parameters for the vector

§Returns

  • The result of the matrix-vector multiplication sequence