Weight Load Time

Problem

Estimate how long it takes to stream a model's weights from memory once: model_bytes / (bandwidth_gbps * 1e6) milliseconds.

The signature is weight_load_time_ms(model_bytes, bandwidth_gbps). Return a float number of milliseconds.

Hint

This is the same memory-bound term that dominates each decode step.

case 1

in: [14000000000,32]

expected: 437.5

case 2

in: [1000000000,1000]

expected: 1

case 3

in: [0,900]

expected: 0

case 4

in: [4000000000,900]

expected: 4.4444444444

Starter codePython

def weight_load_time_ms(model_bytes, bandwidth_gbps):
    # Your code here
    pass

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