Average Bits Per Weight

Problem

Compute the parameter-weighted average bit width across layers: sum(params_i * bits_i) / sum(params_i). An all-zero parameter count returns 0.0.

The signature is average_bits_per_weight(layer_params, layer_bits).

Hint

Mixed-precision reports average more parameters in high precision.

case 1

in: [[1000,3000],[4,16]]

expected: 13

case 2

in: [[1,1],[8,8]]

expected: 8

case 3

in: [[100,0],[4,16]]

expected: 4

case 4

in: [[1,2],[3,6]]

expected: 5

Starter codePython

def average_bits_per_weight(layer_params, layer_bits):
    # Your code here
    pass

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