#deployment

Question 1: Single choice

A model loses too much accuracy when its weights are quantized to 8 bits after training, so you switch to quantization-aware training. What does quantization-aware training do?

It simulates 8-bit rounding in the forward pass while training, so the weights adapt to it1
It quantizes the trained float model once at the end, calibrating ranges on a small dataset2
It trains with 8-bit integer weights, gradients and optimizer states from the start3
It prunes the weights that would round to zero and then retrains the remaining weights4

Timed test