MoE Load Balance

Problem

Compute the expert load imbalance factor: max expert token count divided by the mean count.

The signature is moe_load_balance(expert_counts). Return a float; 1.0 means perfectly balanced.

Hint

Imbalance makes some experts idle while others become throughput bottlenecks.

case 1

in: [[10,10,10,10]]

expected: 1

case 2

in: [[20,0,0,0]]

expected: 4

case 3

in: [[5,5,0,0]]

expected: 2

case 4

in: [[3,1,2,4]]

expected: 1.6

Starter codePython

def moe_load_balance(expert_counts):
    # Your code here
    pass

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