Quality loss with quantization is expected. It seems like with GPTQ the loss is within acceptable range based on the perplexity score shown.