Weight-Only vs Activation Quant

Problem

Compare memory for weight-only versus weight+activation quantization. weight_bytes and activation_bytes are fp16 sizes; return [weight_only_bytes, full_bytes] where weight-only keeps activations at 16 bits and full quantizes activations too.

The signature is weight_only_vs_activation_quant(weight_bytes, activation_bytes, weight_bits, activation_bits).

Hint

Weight-only quantization is easier but leaves activation memory untouched.

Test cases

case 1

in: [1000,200,4,8]

expected: [450,350]

case 2

in: [0,100,4,8]

expected: [100,50]

case 3

in: [1000,0,4,4]

expected: [250,250]

Starter codePython

def weight_only_vs_activation_quant(weight_bytes, activation_bytes, weight_bits, activation_bits):
    # Your code here
    pass

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