My pet peeve - the assumption that quantization is a free lunch. In my experience, it's rarely that simple and often requires tweaks to the model architecture itself. https://www.reddit.com/user/Cultural-Lobster7795