Model Warmup Iterations

Problem

Compute how many warmup iterations are needed for latency to decay to the target: ceil(log(target / initial) / log(decay)). If the initial latency is already within budget, return 0.

The signature is model_warmup_iters(target_latency_ms, initial_latency_ms, decay). Return an integer.

Hint

Warmup lets caches, JIT compilation and clocks stabilize before measurement.

Test cases

case 1

in: [80,200,0.8]

expected: 5

case 2

in: [10,100,0.5]

expected: 4

case 3

in: [100,100,0.9]

expected: 0

case 4

in: [50,200,0.5]

expected: 2

Starter codePython

import math
def model_warmup_iters(target_latency_ms, initial_latency_ms, decay):
    # Your code here
    pass

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