Max Batch Size From Memory

Problem

Find the largest batch that fits the memory left after loading the weights: floor((budget_bytes - model_bytes) / per_sequence_cache_bytes). Return 0 if weights exceed the budget or the per-sequence cache is non-positive.

The signature is max_batch_size(budget_bytes, model_bytes, per_sequence_cache_bytes). Return an integer.

Hint

Floor division gives the number of whole per-sequence caches that fit.

Test cases

case 1

in: [80000000000,40000000000,500000000]

expected: 80

case 2

in: [100000000000,99000000000,1000000000]

expected: 1

case 3

in: [1000000000,2000000000,1000]

expected: 0

case 4

in: [50000000000,0,2500000000]

expected: 20

Starter codePython

def max_batch_size(budget_bytes, model_bytes, per_sequence_cache_bytes):
    # Your code here
    pass

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