Warmup LR

Problem

Linear learning-rate warmup: return base_lr * min(1.0, step / warmup_steps), or base_lr when warmup_steps <= 0.

The signature is warmup_lr(step, warmup_steps, base_lr).

Hint

Clamp the step fraction at 1.0 so the rate plateaus after warmup.

case 1

in: [0,100,0.001]

expected: 0

case 2

in: [50,100,0.001]

expected: 0.0005

case 3

in: [100,100,0.001]

expected: 0.001

case 4

in: [200,100,0.001]

expected: 0.001

case 5

in: [10,0,0.01]

expected: 0.01

Starter codePython

def warmup_lr(step, warmup_steps, base_lr):
    # Your code here
    pass

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