Fake-Quant Forward Value

Problem

Compute the fake-quantization forward value: q = clamp(round(x/scale) + zero_point, qmin, qmax), then dequantize with (q - zero_point) * scale. Apply element-wise to the list x and return the list of dequantized values.

The signature is fake_quant_forward(x, scale, zero_point, qmin, qmax).

Hint

Fake quantization simulates the rounding error while keeping the tensor in floating point.

Test cases

case 1

in: [[1.2,-1.2],1,0,-128,127]

expected: [1,-1]

case 2

in: [[1000],1,0,-128,127]

expected: [127]

case 3

in: [[2.5],0.5,0,0,15]

expected: [2.5]

case 4

in: [[0,5],1,8,0,15]

expected: [0,5]

Starter codePython

def fake_quant_forward(x, scale, zero_point, qmin, qmax):
    # Your code here
    pass

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