modeci_mdf.functions.onnx.qlinearconv

modeci_mdf.functions.onnx.qlinearconv(*args, **kwargs)

The convolution operator consumes a quantized input tensor, its scale and zero point, a quantized filter, its scale and zero point, and output’s scale and zero point, and computes the quantized output. Each scale and zero-point pair must have same shape. It means they must be either scalars (per tensor) or 1-D tensors (per output channel). Each input or output and its related zero point must have same type. When bias is present it must be quantized using scale = input scale * weight scale and zero point as 0.