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())