No, not in milliseconds if you have longish context. Prefill is very compute heavy, compared to inference.