GPTQ models for nn.Linear 4bit weight & 16bit activation (symmetric) integer quantization