Overview > klyn.math.nn.quantization > QuantizedMatrix > multiplyCPU
Version 0.2.2 Overview Index Tree Deprecated
klyn.math.nn.quantization.QuantizedMatrix.multiplyCPU
method
public multiplyCPU(inputs as NDArray<Float>) as NDArray<Float>
Description

Computes inputs times this weight matrix's transpose on the CPU. Inputs and output use Float, with FP32 accumulation. Weights stay packed INT8 throughout; only group dot products are multiplied by their scales.

Parameters
ParameterDescription
inputsFloat matrix with shape [batch, columns].
Returns

Float matrix with shape [batch, rows].

Throws
  • ValueException for incompatible input shapes.
  • ValueException inside a GPU execution scope.
Example

outputs = matrix.multiplyCPU(NDArray<Float>.ones([1l, matrix.columns]))