QAT vs PTQ Error Delta

Problem

Compare post-training quantization (PTQ) and quantization-aware training (QAT) error. Return [ptq_error - qat_error, (ptq_error - qat_error) / ptq_error], using 0.0 for the relative reduction when ptq_error is 0.

The signature is qat_vs_ptq_error(ptq_error, qat_error).

Hint

QAT usually narrows the gap because the network adapts to rounding during fine-tuning.

Test cases

case 1

in: [0.1,0.02]

expected: [0.08,0.7999999999999999]

case 2

in: [0.5,0.5]

expected: [0,0]

case 3

in: [0,0]

expected: [0,0]

case 4

in: [0.3,0.1]

expected: [0.19999999999999998,0.6666666666666666]

Starter codePython

def qat_vs_ptq_error(ptq_error, qat_error):
    # Your code here
    pass

4 test cases · run your code against them in the editor