import os import xnas.core.benchmark as benchmark import xnas.core.config as config import xnas.core.logging as logging import xnas.datasets.loader as loader import xnas.core.distributed as dist from xnas.core.config import cfg logger = logging.get_logger(__name__) def test_full_time(): config.dump_cfg() logging.setup_logging() logger.info("Config:\n{}".format(cfg)) logger.info(logging.dump_log_data(cfg, "cfg")) [train_loader, test_loader] = loader.construct_loader( cfg.SEARCH.DATASET, cfg.SEARCH.SPLIT, cfg.SEARCH.BATCH_SIZE) avg_time = benchmark.compute_full_loader(test_loader, epoch=3) for i, _time in enumerate(avg_time): logger.info("The {}'s epoch average time is: {}".format(i, _time)) if __name__ == "__main__": config.load_cfg_fom_args("Compute model and loader timings.") os.makedirs(cfg.OUT_DIR, exist_ok=True) dist.multi_proc_run(num_proc=1, fun=test_full_time)
import xnas.core.checkpoint as checkpoint import time from torch.utils.tensorboard import SummaryWriter import torch.nn.functional as F import torch.nn as nn import torch import numpy as np import random import os import json import gc import sys sys.path.append(".") # config load and assert config.load_cfg_fom_args() config.assert_and_infer_cfg() cfg.freeze() # tensorboard writer = SummaryWriter(log_dir=os.path.join(cfg.OUT_DIR, "tb")) logger = logging.get_logger(__name__) def random_sampling(search_space, distribution_optimizer, epoch=-1000, _random=False): if _random: num_ops, total_edges = search_space.num_ops, search_space.all_edges
def main(): config.load_cfg_fom_args("Compute model and loader timings.") os.makedirs(cfg.OUT_DIR, exist_ok=True) dist.multi_proc_run(num_proc=1, fun=test_full_time)