DuQuant++

MACaQ quantization method speeds up AI model inference

Researchers have introduced MACaQ, a novel post-training quantization method designed for efficient low-precision neural network inference. This method independently quantizes each linear layer, incorporating online act…