f"--search_teacher_folder {teacher_folder} "
# -----[AUXILIARY]-----
"--aux_store_root store/1.TEST "
"--aux_console_output")
cmd = cmd.strip()
parser = student_search_parser(teacher_type=teacher_type)
args = parser.parse_args(cmd.split(" "))
else: # 2. Running
preset_parser = student_search_preset_parser()
preset_args, unk = preset_parser.parse_known_args()
teacher_type, dataset = preset_args.T, preset_args.D
parser = student_search_parser(teacher_type=teacher_type)
args = parser.parse_args(unk)
assert args.search_teacher_folder is not None
store_folder = setup_folder(store_root=args.aux_store_root,
project_name=args.name)
folder = Path(store_folder)
setup_logger(folder_path=folder, console_output=args.aux_console_output)
setup_seed(args.seed)
if args.gpu_teacher == args.gpu_student:
gpu_str = args.gpu_teacher
device_student = torch.device("cuda:0")
device_teacher = torch.device("cuda:0")
else: # args.gpu_teacher != args.gpu_student:
if args.gpu_teacher > args.gpu_student:
gpu_str = f"{args.gpu_student},{args.gpu_teacher}"
device_student = torch.device("cuda:0")
device_teacher = torch.device("cuda:1")
else: # args.gpu_teacher < args.gpu_student:
data["block_shape"] = [int(t) for t in blocks.split(",")]
data.update(patch)
return data
if __name__ == "__main__":
# DEBUG setup
pack = {
# "console_output": True,
# "log_every": 20,
}
configs = parse_arg(**pack)
# 0. setup folder and logger
path = setup_folder(configs)
configs.update({"store_path": path})
logger.setup_logger(configs)
# 1. setup env, seed and gpu
environment_preset(configs)
gpu_config = get_gpu_config(configs["occupy"])
sess = tf.Session(config=gpu_config)
# 2. prepare dataset
data_loader = DataLoaderPretrain(configs, configs["min_freq"])
train_set, test_set = data_loader.split()
data_loader.save_dict()
logging.info("item dict saved to itemdict.pkl")
# 3. update model configs