MoE Throughput Gain

Problem

Compute the sparse-activation throughput gain of a mixture-of-experts model: total_params / active_params.

The signature is moe_throughput_gain(total_params, active_params).

Hint

Sparse MoE scales total parameters without scaling per-token compute.

case 1

in: [1000000,100000]

expected: 10

case 2

in: [8,8]

expected: 1

case 3

in: [47000000000,13000000000]

expected: 3.6153846154

case 4

in: [100,10]

expected: 10

Starter codePython

def moe_throughput_gain(total_params, active_params):
    # Your code here
    pass

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