Why Some Models Quantize Better Than Others / hacker news