|
- """
- Based on Deit: Facebook, Inc.
- https://github.com/facebookresearch/deit/blob/main/main.py
- """
-
- import argparse
- import datetime
- import time
- import os
- import torch.backends.cudnn as cudnn
- import json
- import torch
- import numpy as np
- from pathlib import Path
-
- from timm.data import Mixup
- from timm.loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy
- from timm.scheduler import create_scheduler
- from timm.optim import create_optimizer
- #from timm.utils import get_state_dict
- from timm.utils import NativeScaler
-
- from datasets import build_dataset
- from engine import train_one_epoch, evaluate
- from samplers import RASampler
- from models import *
- import utils
- from swin_models import *
- from mobilenetv2 import my_mobilenet_v2,timm_mobilenet_v2, vis_mobilenet_v2
-
- def get_args_parser():
- parser = argparse.ArgumentParser('DeiT training and evaluation script', add_help=False)
- parser.add_argument('--batch-size', default=64, type=int)
- parser.add_argument('--epochs', default=3, type=int)
-
- # Model parameters
- parser.add_argument('--model', default='deit_base_patch16_224', type=str, metavar='MODEL',
- help='Name of model to train')
- parser.add_argument('--input-size', default=224, type=int, help='images input size')
-
- parser.add_argument('--drop', type=float, default=0.0, metavar='PCT',
- help='Dropout rate (default: 0.)')
- parser.add_argument('--drop-path', type=float, default=0.1, metavar='PCT',
- help='Drop path rate (default: 0.1)')
-
- parser.add_argument('--qk-scale-factor', type=float, default=None, metavar='PCT',
- help='scale q & k in self-attention. scale = head_dim ** qk-scale-factor')
-
- # Optimizer parameters
- parser.add_argument('--amp', action='store_true', default=False,
- help='using automatic mixed precision (amp)')
- parser.add_argument('--opt', default='adamw', type=str, metavar='OPTIMIZER',
- help='Optimizer (default: "adamw"')
- parser.add_argument('--opt-eps', default=1e-8, type=float, metavar='EPSILON',
- help='Optimizer Epsilon (default: 1e-8)')
- parser.add_argument('--opt-betas', default=None, type=float, nargs='+', metavar='BETA',
- help='Optimizer Betas (default: None, use opt default)')
- parser.add_argument('--clip-grad', type=float, default=None, metavar='NORM',
- help='Clip gradient norm (default: None, no clipping)')
- parser.add_argument('--momentum', type=float, default=0.9, metavar='M',
- help='SGD momentum (default: 0.9)')
- parser.add_argument('--weight-decay', type=float, default=0.05,
- help='weight decay (default: 0.05)')
- # Learning rate schedule parameters
- parser.add_argument('--sched', default='cosine', type=str, metavar='SCHEDULER',
- help='LR scheduler (default: "cosine"')
- parser.add_argument('--lr', type=float, default=5e-4, metavar='LR',
- help='learning rate (default: 5e-4)')
- parser.add_argument('--warmup-lr', type=float, default=1e-6, metavar='LR',
- help='warmup learning rate (default: 1e-6)')
- parser.add_argument('--min-lr', type=float, default=1e-5, metavar='LR',
- help='lower lr bound for cyclic schedulers that hit 0 (1e-5)')
-
- parser.add_argument('--decay-epochs', type=float, default=0, metavar='N',
- help='epoch interval to decay LR')
- parser.add_argument('--warmup-epochs', type=int, default=0, metavar='N',
- help='epochs to warmup LR, if scheduler supports')
-
- parser.add_argument('--cooldown-epochs', type=int, default=10, metavar='N',
- help='epochs to cooldown LR at min_lr, after cyclic schedule ends')
- #parser.add_argument('--patience-epochs', type=int, default=10, metavar='N',
- # help='patience epochs for Plateau LR scheduler (default: 10')
- parser.add_argument('--decay-rate', '--dr', type=float, default=0.1, metavar='RATE',
- help='LR decay rate (default: 0.1)')
- parser.add_argument('--skip_test', action='store_true', default=False,
- help='skip 250 eval epoch')
-
- # Augmentation parameters
- parser.add_argument('--color-jitter', type=float, default=0.4, metavar='PCT',
- help='Color jitter factor (default: 0.4)')
- parser.add_argument('--aa', type=str, default='rand-m9-mstd0.5-inc1', metavar='NAME',
- help='Use AutoAugment policy. "v0" or "original". " + \
- "(default: rand-m9-mstd0.5-inc1)'),
- parser.add_argument('--smoothing', type=float, default=0.1, help='Label smoothing (default: 0.1)')
- parser.add_argument('--train-interpolation', type=str, default='bicubic',
- help='Training interpolation (random, bilinear, bicubic default: "bicubic")')
-
- parser.add_argument('--std-aug', action='store_true', default=False)
- parser.add_argument('--repeated-aug', action='store_true')
- parser.add_argument('--no-repeated-aug', action='store_false', dest='repeated_aug')
- parser.set_defaults(repeated_aug=True)
-
- # * Random Erase params
- parser.add_argument('--reprob', type=float, default=0.25, metavar='PCT',
- help='Random erase prob (default: 0.25)')
- parser.add_argument('--remode', type=str, default='pixel',
- help='Random erase mode (default: "pixel")')
- parser.add_argument('--recount', type=int, default=1,
- help='Random erase count (default: 1)')
- parser.add_argument('--resplit', action='store_true', default=False,
- help='Do not random erase first (clean) augmentation split')
-
- # * Mixup params
- parser.add_argument('--mixup', type=float, default=0.8,
- help='mixup alpha, mixup enabled if > 0. (default: 0.8)')
- parser.add_argument('--cutmix', type=float, default=1.0,
- help='cutmix alpha, cutmix enabled if > 0. (default: 1.0)')
- parser.add_argument('--cutmix-minmax', type=float, nargs='+', default=None,
- help='cutmix min/max ratio, overrides alpha and enables cutmix if set (default: None)')
- parser.add_argument('--mixup-prob', type=float, default=1.0,
- help='Probability of performing mixup or cutmix when either/both is enabled')
- parser.add_argument('--mixup-switch-prob', type=float, default=0.5,
- help='Probability of switching to cutmix when both mixup and cutmix enabled')
- parser.add_argument('--mixup-mode', type=str, default='batch',
- help='How to apply mixup/cutmix params. Per "batch", "pair", or "elem"')
-
- # Dataset parameters
- parser.add_argument('--data-path', default='/dataset/imagenet/', type=str,
- help='dataset path')
- parser.add_argument('--data-set', default='IMNET', choices=['CIFAR', 'IMNET', 'IMNET100', 'IMNET10'],
- type=str, help='Image Net dataset path')
-
- parser.add_argument('--output_dir', default='/model',
- help='path where to save, empty for no saving')
- parser.add_argument('--device', default='cuda',
- help='device to use for training / testing')
- parser.add_argument('--seed', default=0, type=int)
- parser.add_argument('--resume', default='', help='resume from checkpoint')
- parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
- help='start epoch')
- parser.add_argument('--eval', action='store_true', help='Perform evaluation only')
- parser.add_argument('--dist-eval', action='store_true', default=False, help='Enabling distributed evaluation')
- parser.add_argument('--num_workers', default=10, type=int)
- parser.add_argument('--pin-mem', action='store_true',
- help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
- parser.add_argument('--no-pin-mem', action='store_false', dest='pin_mem',
- help='')
- parser.set_defaults(pin_mem=True)
-
- # distributed training parameters
- parser.add_argument('--world_size', default=1, type=int,
- help='number of distributed processes')
- parser.add_argument('--dist_url', default='env://', help='url used to set up distributed training')
- return parser
-
- def main(args):
- utils.init_distributed_mode(args)
-
- print(args)
-
- device = torch.device(args.device)
-
- # fix the seed for reproducibility
- seed = args.seed + utils.get_rank()
- torch.manual_seed(seed)
- np.random.seed(seed)
- #random.seed(seed)
-
- cudnn.benchmark = True
-
- dataset_train, args.nb_classes = build_dataset(is_train=True, args=args)
- dataset_val, _ = build_dataset(is_train=False, args=args)
-
- if args.distributed:
- num_tasks = utils.get_world_size()
- global_rank = utils.get_rank()
- if args.repeated_aug:
- sampler_train = RASampler(
- dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
- )
- else:
- sampler_train = torch.utils.data.DistributedSampler(
- dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
- )
- if args.dist_eval:
- if len(dataset_val) % num_tasks != 0:
- print('Warning: Enabling distributed evaluation with an eval dataset not divisible by process number. '
- 'This will slightly alter validation results as extra duplicate entries are added to achieve '
- 'equal num of samples per-process.')
- sampler_val = torch.utils.data.DistributedSampler(
- dataset_val, num_replicas=num_tasks, rank=global_rank, shuffle=False)
- else:
- sampler_val = torch.utils.data.SequentialSampler(dataset_val)
- else:
- sampler_train = torch.utils.data.RandomSampler(dataset_train)
- sampler_val = torch.utils.data.SequentialSampler(dataset_val)
-
- data_loader_train = torch.utils.data.DataLoader(
- dataset_train, sampler=sampler_train,
- batch_size=args.batch_size,
- num_workers=args.num_workers,
- pin_memory=args.pin_mem,
- drop_last=True,
- )
-
- data_loader_val = torch.utils.data.DataLoader(
- dataset_val, sampler=sampler_val,
- batch_size=int(1.5 * args.batch_size),
- num_workers=args.num_workers,
- pin_memory=args.pin_mem,
- drop_last=False
- )
-
- mixup_fn = None
- mixup_active = args.mixup > 0 or args.cutmix > 0. or args.cutmix_minmax is not None
- if mixup_active:
- mixup_fn = Mixup(
- mixup_alpha=args.mixup, cutmix_alpha=args.cutmix, cutmix_minmax=args.cutmix_minmax,
- prob=args.mixup_prob, switch_prob=args.mixup_switch_prob, mode=args.mixup_mode,
- label_smoothing=args.smoothing, num_classes=args.nb_classes)
-
- # print(f"Creating model: {args.model}")
-
- if 'mobile' in args.model:
- model = eval(args.model+'()')
- else:
- model = eval(args.model)(
- num_classes=args.nb_classes,
- drop_rate=args.drop,
- drop_path_rate=args.drop_path,
- qk_scale=args.qk_scale_factor
- )
-
- model.to(device)
-
- model_without_ddp = model
- if args.distributed:
- model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
- model_without_ddp = model.module
- n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
- print('number of params:', n_parameters)
-
- linear_scaled_lr = args.lr * args.batch_size * utils.get_world_size() / 512.0
- args.lr = linear_scaled_lr
- optimizer = create_optimizer(args, model_without_ddp)
- loss_scaler = NativeScaler()
- lr_scheduler, _ = create_scheduler(args, optimizer)
-
- #criterion = LabelSmoothingCrossEntropy()
-
- if args.mixup > 0.:
- # smoothing is handled with mixup label transform
- criterion = SoftTargetCrossEntropy()
- elif args.smoothing:
- criterion = LabelSmoothingCrossEntropy(smoothing=args.smoothing)
- else:
- criterion = torch.nn.CrossEntropyLoss()
-
- output_dir = Path(args.output_dir)
- if args.resume:
- if args.resume.startswith('https'):
- checkpoint = torch.hub.load_state_dict_from_url(
- args.resume, map_location='cpu', check_hash=True)
- else:
- checkpoint = torch.load(args.resume, map_location='cpu')
- if 'model' in checkpoint:
- model_without_ddp.load_state_dict(checkpoint['model'])
- else:
- model_without_ddp.load_state_dict(checkpoint)
- if not args.eval and 'optimizer' in checkpoint and 'lr_scheduler' in checkpoint and 'epoch' in checkpoint:
- optimizer.load_state_dict(checkpoint['optimizer'])
- lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
- args.start_epoch = checkpoint['epoch'] + 1
- if 'scaler' in checkpoint:
- loss_scaler.load_state_dict(checkpoint['scaler'])
-
- if args.eval:
- test_stats = evaluate(data_loader_val, model, device, amp=args.amp)
- # print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']}")
- return
-
- print(f"Start training for {args.epochs} epochs")
- start_time = time.time()
- max_accuracy = 0.0
- for epoch in range(args.start_epoch, args.epochs):
- if args.distributed:
- data_loader_train.sampler.set_epoch(epoch)
-
- train_stats = train_one_epoch(
- model, criterion, data_loader_train,
- optimizer, device, epoch, loss_scaler, mixup_fn, amp=args.amp
- )
-
- lr_scheduler.step(epoch)
- if args.output_dir:
- checkpoint_paths = [output_dir / 'checkpoint.pth']
- for checkpoint_path in checkpoint_paths:
- utils.save_on_master({
- 'model': model_without_ddp.state_dict(),
- 'optimizer': optimizer.state_dict(),
- 'lr_scheduler': lr_scheduler.state_dict(),
- 'epoch': epoch,
- #'model_ema': get_state_dict(model_ema),
- 'scaler': loss_scaler.state_dict(),
- 'args': args,
- }, checkpoint_path)
-
- if args.skip_test is False or epoch >= 250:
-
- test_stats = evaluate(data_loader_val, model, device, amp=args.amp)
- # print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']}")
- if test_stats["acc1"] > max_accuracy and args.output_dir is not None:
- checkpoint_paths = [output_dir / 'best.pth']
- for checkpoint_path in checkpoint_paths:
- utils.save_on_master({
- 'model': model_without_ddp.state_dict(),
- 'optimizer': optimizer.state_dict(),
- 'lr_scheduler': lr_scheduler.state_dict(),
- 'epoch': epoch,
- #'model_ema': get_state_dict(model_ema),
- 'scaler': loss_scaler.state_dict(),
- 'args': args,
- }, checkpoint_path)
- max_accuracy = max(max_accuracy, test_stats["acc1"])
- print(f'Max accuracy: {max_accuracy}')
-
- log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
- **{f'test_{k}': v for k, v in test_stats.items()},
- 'epoch': epoch,
- 'n_parameters': n_parameters}
-
- if args.output_dir and utils.is_main_process():
- with (output_dir / "log.txt").open("a") as f:
- f.write(json.dumps(log_stats) + "\n")
-
- total_time = time.time() - start_time
- total_time_str = str(datetime.timedelta(seconds=int(total_time)))
- print('Training time {}'.format(total_time_str))
-
-
- def C2netMultiObsToEnv(multi_data_url, data_dir):
- #--multi_data_url is json data, need to do json parsing for multi_data_url
- print(multi_data_url)
- multi_data_json = json.loads(multi_data_url)
- for i in range(len(multi_data_json)):
- zipfile_path = data_dir + "/" + multi_data_json[i]["dataset_name"]
- try:
- mox.file.copy(multi_data_json[i]["dataset_url"], zipfile_path)
- print("Successfully Download {} to {}".format(multi_data_json[i]["dataset_url"],zipfile_path))
- #get filename and unzip the dataset
- filename = os.path.splitext(multi_data_json[i]["dataset_name"])[0]
- filePath = data_dir + "/" + filename
- if not os.path.exists(filePath):
- os.makedirs(filePath)
- os.system("unzip {} -d {}".format(zipfile_path, filePath))
-
- except Exception as e:
- print('moxing download {} to {} failed: '.format(
- multi_data_json[i]["dataset_url"], zipfile_path) + str(e))
- #Set a cache file to determine whether the data has been copied to obs.
- #If this file exists during multi-card training, there is no need to copy the dataset multiple times.
- f = open("/cache/download_input.txt", 'w')
- f.close()
- try:
- if os.path.exists("/cache/download_input.txt"):
- print("download_input succeed")
- except Exception as e:
- print("download_input failed")
- return
-
-
- def DownloadFromQizhi(multi_data_url, data_dir):
- device_num = int(os.getenv('RANK_SIZE'))
- if device_num == 1:
- C2netMultiObsToEnv(multi_data_url,data_dir)
- context.set_context(mode=context.GRAPH_MODE,device_target=args.device_target)
- if device_num > 1:
- # set device_id and init for multi-card training
- context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target, device_id=int(os.getenv('ASCEND_DEVICE_ID')))
- context.reset_auto_parallel_context()
- context.set_auto_parallel_context(device_num = device_num, parallel_mode=ParallelMode.DATA_PARALLEL, gradients_mean=True, parameter_broadcast=True)
- init()
- #Copying obs data does not need to be executed multiple times, just let the 0th card copy the data
- local_rank=int(os.getenv('RANK_ID'))
- if local_rank%8==0:
- C2netMultiObsToEnv(multi_data_url,data_dir)
- #If the cache file does not exist, it means that the copy data has not been completed,
- #and Wait for 0th card to finish copying data
- while not os.path.exists("/cache/download_input.txt"):
- time.sleep(1)
- return
-
-
- def EnvToObs(train_dir, obs_train_url):
- try:
- mox.file.copy_parallel(train_dir, obs_train_url)
- print("Successfully Upload {} to {}".format(train_dir,
- obs_train_url))
- except Exception as e:
- print('moxing upload {} to {} failed: '.format(train_dir,
- obs_train_url) + str(e))
- return
-
-
- def UploadToQizhi(train_dir, obs_train_url):
- device_num = int(os.getenv('RANK_SIZE'))
- local_rank=int(os.getenv('RANK_ID'))
- if device_num == 1:
- EnvToObs(train_dir, obs_train_url)
- if device_num > 1:
- if local_rank%8==0:
- EnvToObs(train_dir, obs_train_url)
- return
-
-
- if __name__ == "__main__":
- parser = argparse.ArgumentParser('DeiT training and evaluation script', parents=[get_args_parser()])
- args = parser.parse_args()
- subdir = os.listdir("/dataset");
- for sub in subdir:
- print(sub)
-
- if args.output_dir:
- Path(args.output_dir).mkdir(parents=True, exist_ok=True)
- main(args)
|