ParentFull threadfzliu·AFAIK for over-parameterized models, performing quantization or any other form of compression won't reduce accuracy by much (don't quote me on this though).View on HN