benchmarks/serialization/nested_annotation_str.py

import torch
import torch.utils.benchmark as benchmark

MEMO = {}
def create_nested_dict_type(layers):
    if layers == 0:
        return torch._C.StringType.get()
    if layers not in MEMO:
        less_nested = create_nested_dict_type(layers - 1)
        result = torch._C.DictType(torch._C.StringType.get(), torch._C.TupleType([less_nested, less_nested]))
        MEMO[layers] = result
    return MEMO[layers]


nesting_levels = (1, 3, 5, 10)
types = (reasonable, medium, big, huge) = [create_nested_dict_type(x) for x in nesting_levels]

timers = [benchmark.Timer(stmt='x.annotation_str', globals={'x': nested_type}) for nested_type in types]

for nesting_level, typ, timer in zip(nesting_levels, types, timers):
    print("Nesting level:", nesting_level)
    print("output:", typ.annotation_str[:70])
    print(timer.blocked_autorange())