Comment by WanderPanda

I find it hard to trust post training quantizations. Why don't they run benchmarks to see the degradation in performance? It sketches me out because it should be the easiest thing to automatically run a suite of benchmarks