I am using auto3dseg for segmentation experiments, my running script is

from monai.apps.auto3dseg import AutoRunner

if __name__ == '__main__':
    runner = AutoRunner(input={"name": "Task500_XXX",
                                "task": "segmentation",
                                "modality": "CT",
                                "datalist": "/home/XXX/Projects/MONAI/dataset.json",
                                "dataroot": "/media/DATA2/XXX/20221124/",
                                "class_names": ["upper", "lower"]},
                        work_dir="/home/XXX/Projects/MONAI/work_dir",
                        analyze=True,
                        algo_gen=True)
    runner.run()

First, there is no problem with generating dataset.yaml by dataset analysis, and then there is no problem with generating algorithm. Next, start training and run the file work_dir/dints_0/scripts/search.py
After a period of time, the machine has a memory overflow and the program is killed (my machine has 128G memory)
After looking at the code, I suspect that it is caused by data caching
https://github.com/Project-MONAI/research-contributions/blob/1dde7a112c69785b4fb5cd58d5b73f3fa5bdc577/auto3dseg/algorithm_templates/dints/scripts/search.py#L168
My data has about 300 copies of CT data, and they cannot all be cached.
I want to know how to modify this parameter, I tried to modify input.yaml or runner.set_train_param() ,but it didn’t work