DL-Art-School/codes/train.py

370 lines
18 KiB
Python
Raw Normal View History

2019-08-23 13:42:47 +00:00
import os
import math
import argparse
import random
import logging
import shutil
2020-04-22 06:38:53 +00:00
from tqdm import tqdm
2019-08-23 13:42:47 +00:00
import torch
from data.data_sampler import DistIterSampler
2020-12-31 03:18:58 +00:00
from trainer.eval.evaluator import create_evaluator
2019-08-23 13:42:47 +00:00
from utils import util, options as option
from data import create_dataloader, create_dataset, get_dataset_debugger
2020-12-18 16:18:34 +00:00
from trainer.ExtensibleTrainer import ExtensibleTrainer
2020-06-18 17:28:55 +00:00
from time import time
from datetime import datetime
2019-08-23 13:42:47 +00:00
2022-01-24 22:08:13 +00:00
from utils.util import opt_get, map_cuda_to_correct_device
def init_dist(backend, **kwargs):
# These packages have globals that screw with Windows, so only import them if needed.
import torch.distributed as dist
2022-03-04 19:46:41 +00:00
rank = int(os.environ['LOCAL_RANK'])
assert rank < torch.cuda.device_count()
torch.cuda.set_device(rank)
dist.init_process_group(backend=backend, **kwargs)
class Trainer:
def init(self, opt_path, opt, launcher):
self._profile = False
self.val_compute_psnr = opt_get(opt, ['eval', 'compute_psnr'], False)
self.val_compute_fea = opt_get(opt, ['eval', 'compute_fea'], False)
self.current_step = 0
self.total_training_data_encountered = 0
#### loading resume state if exists
if opt['path'].get('resume_state', None):
# distributed resuming: all load into default GPU
2022-01-24 22:08:13 +00:00
resume_state = torch.load(opt['path']['resume_state'], map_location=map_cuda_to_correct_device)
else:
resume_state = None
#### mkdir and loggers
if self.rank <= 0: # normal training (self.rank -1) OR distributed training (self.rank 0)
if resume_state is None:
util.mkdir_and_rename(
opt['path']['experiments_root']) # rename experiment folder if exists
util.mkdirs(
(path for key, path in opt['path'].items() if not key == 'experiments_root' and path is not None
and 'pretrain_model' not in key and 'resume' not in key))
shutil.copy(opt_path, os.path.join(opt['path']['experiments_root'], f'{datetime.now().strftime("%d%m%Y_%H%M%S")}_{os.path.basename(opt_path)}'))
# config loggers. Before it, the log will not work
util.setup_logger('base', opt['path']['log'], 'train_' + opt['name'], level=logging.INFO,
screen=True, tofile=True)
self.logger = logging.getLogger('base')
self.logger.info(option.dict2str(opt))
# tensorboard logger
if opt['use_tb_logger'] and 'debug' not in opt['name']:
self.tb_logger_path = os.path.join(opt['path']['experiments_root'], 'tb_logger')
from torch.utils.tensorboard import SummaryWriter
self.tb_logger = SummaryWriter(log_dir=self.tb_logger_path)
else:
util.setup_logger('base', opt['path']['log'], 'train', level=logging.INFO, screen=True)
self.logger = logging.getLogger('base')
if resume_state is not None:
option.check_resume(opt, resume_state['iter']) # check resume options
# convert to NoneDict, which returns None for missing keys
opt = option.dict_to_nonedict(opt)
self.opt = opt
2020-11-12 22:45:25 +00:00
#### wandb init
2021-06-06 22:52:07 +00:00
if opt['wandb'] and self.rank <= 0:
2020-11-12 22:45:25 +00:00
import wandb
os.makedirs(os.path.join(opt['path']['log'], 'wandb'), exist_ok=True)
2021-11-22 23:40:05 +00:00
project_name = opt_get(opt, ['wandb_project_name'], opt['name'])
2021-11-23 00:31:29 +00:00
run_name = opt_get(opt, ['wandb_run_name'], None)
wandb.init(project=project_name, dir=opt['path']['log'], config=opt, name=run_name)
2020-11-12 22:45:25 +00:00
#### random seed
seed = opt['train']['manual_seed']
if seed is None:
seed = random.randint(1, 10000)
if self.rank <= 0:
self.logger.info('Random seed: {}'.format(seed))
seed += self.rank # Different multiprocessing instances should behave differently.
util.set_random_seed(seed)
torch.backends.cudnn.benchmark = opt_get(opt, ['cuda_benchmarking_enabled'], True)
# torch.backends.cudnn.deterministic = True
if opt_get(opt, ['anomaly_detection'], False):
torch.autograd.set_detect_anomaly(True)
# Save the compiled opt dict to the global loaded_options variable.
util.loaded_options = opt
#### create train and val dataloader
dataset_ratio = 1 # enlarge the size of each epoch
for phase, dataset_opt in opt['datasets'].items():
if phase == 'train':
self.train_set, collate_fn = create_dataset(dataset_opt, return_collate=True)
self.dataset_debugger = get_dataset_debugger(dataset_opt)
if self.dataset_debugger is not None and resume_state is not None:
self.dataset_debugger.load_state(opt_get(resume_state, ['dataset_debugger_state'], {}))
train_size = int(math.ceil(len(self.train_set) / dataset_opt['batch_size']))
total_iters = int(opt['train']['niter'])
self.total_epochs = int(math.ceil(total_iters / train_size))
if opt['dist']:
self.train_sampler = DistIterSampler(self.train_set, self.world_size, self.rank, dataset_ratio)
self.total_epochs = int(math.ceil(total_iters / (train_size * dataset_ratio)))
2021-08-19 22:45:34 +00:00
shuffle = False
else:
2020-10-27 21:24:05 +00:00
self.train_sampler = None
2021-08-19 22:45:34 +00:00
shuffle = True
self.train_loader = create_dataloader(self.train_set, dataset_opt, opt, self.train_sampler, collate_fn=collate_fn, shuffle=shuffle)
if self.rank <= 0:
2022-05-02 06:11:26 +00:00
self.logger.info('Number of training data elements: {:,d}, iters: {:,d}'.format(
len(self.train_set), train_size))
self.logger.info('Total epochs needed: {:d} for iters {:,d}'.format(
self.total_epochs, total_iters))
elif phase == 'val':
self.val_set, collate_fn = create_dataset(dataset_opt, return_collate=True)
self.val_loader = create_dataloader(self.val_set, dataset_opt, opt, None, collate_fn=collate_fn)
if self.rank <= 0:
self.logger.info('Number of val images in [{:s}]: {:d}'.format(
dataset_opt['name'], len(self.val_set)))
2020-10-22 19:27:32 +00:00
else:
raise NotImplementedError('Phase [{:s}] is not recognized.'.format(phase))
assert self.train_loader is not None
#### create model
self.model = ExtensibleTrainer(opt)
### Evaluators
self.evaluators = []
if 'eval' in opt.keys() and 'evaluators' in opt['eval'].keys():
# In "pure" mode, we propagate through the normal training steps, but use validation data instead and average
# the total loss. A validation dataloader is required.
if opt_get(opt, ['eval', 'pure'], False):
assert hasattr(self, 'val_loader')
for ev_key, ev_opt in opt['eval']['evaluators'].items():
self.evaluators.append(create_evaluator(self.model.networks[ev_opt['for']],
ev_opt, self.model.env))
#### resume training
if resume_state:
self.logger.info('Resuming training from epoch: {}, iter: {}.'.format(
resume_state['epoch'], resume_state['iter']))
self.start_epoch = resume_state['epoch']
self.current_step = resume_state['iter']
self.total_training_data_encountered = opt_get(resume_state, ['total_data_processed'], 0)
2022-05-23 16:54:37 +00:00
if opt_get(opt, ['path', 'optimizer_reset'], False):
print('!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!')
print('!! RESETTING OPTIMIZER STATES')
print('!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!')
else:
self.model.resume_training(resume_state, 'amp_opt_level' in opt.keys()) # handle optimizers and schedulers
2020-10-22 19:27:32 +00:00
else:
self.current_step = -1 if 'start_step' not in opt.keys() else opt['start_step']
self.total_training_data_encountered = 0 if 'training_data_encountered' not in opt.keys() else opt['training_data_encountered']
self.start_epoch = 0
if 'force_start_step' in opt.keys():
self.current_step = opt['force_start_step']
self.total_training_data_encountered = self.current_step * opt['datasets']['train']['batch_size']
2021-01-03 05:24:12 +00:00
opt['current_step'] = self.current_step
2022-04-07 17:29:57 +00:00
#### validation
if 'val_freq' in opt['train'].keys():
self.val_freq = opt['train']['val_freq'] * opt['datasets']['train']['batch_size']
else:
self.val_freq = int(opt['train']['val_freq_megasamples'] * 1000000)
2022-06-03 17:20:09 +00:00
self.next_eval_step = self.total_training_data_encountered + self.val_freq
del resume_state # For whatever reason, this relieves a memory burden on the first GPU for some training sessions.
2022-04-07 17:29:57 +00:00
def do_step(self, train_data):
if self._profile:
print("Data fetch: %f" % (time() - _t))
_t = time()
opt = self.opt
batch_size = self.opt['datasets']['train']['batch_size'] # It may seem weird to derive this from opt, rather than train_data. The reason this is done is
# because train_data is process-local while the opt variant represents all of the data fed across all GPUs.
self.current_step += 1
self.total_training_data_encountered += batch_size
2022-03-02 03:19:52 +00:00
will_log = self.current_step % opt['logger']['print_freq'] == 0
#### update learning rate
self.model.update_learning_rate(self.current_step, warmup_iter=opt['train']['warmup_iter'])
#### training
if self._profile:
print("Update LR: %f" % (time() - _t))
2022-04-04 18:33:25 +00:00
_t = time()
2020-11-12 22:45:25 +00:00
self.model.feed_data(train_data, self.current_step)
2022-03-02 03:19:52 +00:00
gradient_norms_dict = self.model.optimize_parameters(self.current_step, return_grad_norms=will_log)
2022-04-04 18:33:25 +00:00
iteration_rate = (time() - _t) / batch_size
if self._profile:
print("Model feed + step: %f" % (time() - _t))
_t = time()
#### log
if self.dataset_debugger is not None:
self.dataset_debugger.update(train_data)
if will_log:
# Must be run by all instances to gather consensus.
current_model_logs = self.model.get_current_log(self.current_step)
2022-03-02 03:19:52 +00:00
if will_log and self.rank <= 0:
logs = {'step': self.current_step,
'samples': self.total_training_data_encountered,
2022-04-04 18:33:25 +00:00
'megasamples': self.total_training_data_encountered / 1000000,
'iteration_rate': iteration_rate}
logs.update(current_model_logs)
if self.dataset_debugger is not None:
logs.update(self.dataset_debugger.get_debugging_map())
2022-03-02 03:19:52 +00:00
logs.update(gradient_norms_dict)
message = '[epoch:{:3d}, iter:{:8,d}, lr:('.format(self.epoch, self.current_step)
for v in self.model.get_current_learning_rate():
message += '{:.3e},'.format(v)
message += ')] '
for k, v in logs.items():
if 'histogram' in k:
self.tb_logger.add_histogram(k, v, self.current_step)
elif isinstance(v, dict):
self.tb_logger.add_scalars(k, v, self.current_step)
else:
message += '{:s}: {:.4e} '.format(k, v)
2019-08-23 13:42:47 +00:00
# tensorboard logger
if opt['use_tb_logger'] and 'debug' not in opt['name']:
self.tb_logger.add_scalar(k, v, self.current_step)
2021-06-06 22:52:07 +00:00
if opt['wandb'] and self.rank <= 0:
2020-11-12 04:48:56 +00:00
import wandb
2022-05-30 22:44:39 +00:00
wandb_logs = {}
for k, v in logs.items():
if 'histogram' in k:
wandb_logs[k] = wandb.Histogram(v)
else:
wandb_logs[k] = v
if opt_get(opt, ['wandb_progress_use_raw_steps'], False):
2022-05-30 22:48:18 +00:00
wandb.log(wandb_logs, step=self.current_step)
else:
2022-05-30 22:48:18 +00:00
wandb.log(wandb_logs, step=self.total_training_data_encountered)
self.logger.info(message)
#### save models and training states
if self.current_step % opt['logger']['save_checkpoint_freq'] == 0:
2022-01-25 01:12:08 +00:00
self.model.consolidate_state()
if self.rank <= 0:
self.logger.info('Saving models and training states.')
self.model.save(self.current_step)
state = {'epoch': self.epoch, 'iter': self.current_step, 'total_data_processed': self.total_training_data_encountered}
if self.dataset_debugger is not None:
state['dataset_debugger_state'] = self.dataset_debugger.get_state()
self.model.save_training_state(state)
if 'alt_path' in opt['path'].keys():
import shutil
print("Synchronizing tb_logger to alt_path..")
alt_tblogger = os.path.join(opt['path']['alt_path'], "tb_logger")
shutil.rmtree(alt_tblogger, ignore_errors=True)
shutil.copytree(self.tb_logger_path, alt_tblogger)
2022-04-07 17:29:57 +00:00
do_eval = self.total_training_data_encountered > self.next_eval_step
if do_eval:
self.next_eval_step = self.total_training_data_encountered + self.val_freq
if opt_get(opt, ['eval', 'pure'], False) and do_eval:
metrics = []
for val_data in tqdm(self.val_loader):
2021-08-09 22:02:01 +00:00
self.model.feed_data(val_data, self.current_step, perform_micro_batching=False)
metrics.append(self.model.test())
reduced_metrics = {}
for metric in metrics:
for k, v in metric.as_dict().items():
if isinstance(v, torch.Tensor) and len(v.shape) == 0:
if k in reduced_metrics.keys():
reduced_metrics[k].append(v)
else:
reduced_metrics[k] = [v]
if self.rank <= 0:
for k, v in reduced_metrics.items():
val = torch.stack(v).mean().item()
2021-08-15 15:09:51 +00:00
self.tb_logger.add_scalar(f'val_{k}', val, self.current_step)
print(f">>Eval {k}: {val}")
if opt['wandb']:
import wandb
wandb.log({f'eval_{k}': torch.stack(v).mean().item() for k,v in reduced_metrics.items()})
2022-04-07 17:29:57 +00:00
if len(self.evaluators) != 0 and do_eval:
eval_dict = {}
for eval in self.evaluators:
2021-06-14 15:51:44 +00:00
if eval.uses_all_ddp or self.rank <= 0:
eval_dict.update(eval.perform_eval())
if self.rank <= 0:
print("Evaluator results: ", eval_dict)
for ek, ev in eval_dict.items():
self.tb_logger.add_scalar(ek, ev, self.current_step)
2021-06-05 05:23:20 +00:00
if opt['wandb']:
2021-06-05 05:27:15 +00:00
import wandb
2021-06-05 05:23:20 +00:00
wandb.log(eval_dict)
2022-01-09 03:31:19 +00:00
# Should not be necessary, but make absolutely sure that there is no grad leakage from validation runs.
for net in self.model.networks.values():
net.zero_grad()
def do_training(self):
self.logger.info('Start training from epoch: {:d}, iter: {:d}'.format(self.start_epoch, self.current_step))
for epoch in range(self.start_epoch, self.total_epochs + 1):
self.epoch = epoch
2022-02-11 18:18:15 +00:00
if self.opt['dist']:
self.train_sampler.set_epoch(epoch)
2021-12-28 23:18:12 +00:00
tq_ldr = tqdm(self.train_loader) if self.rank <= 0 else self.train_loader
_t = time()
for train_data in tq_ldr:
self.do_step(train_data)
def create_training_generator(self, index):
self.logger.info('Start training from epoch: {:d}, iter: {:d}'.format(self.start_epoch, self.current_step))
for epoch in range(self.start_epoch, self.total_epochs + 1):
self.epoch = epoch
if self.opt['dist']:
self.train_sampler.set_epoch(epoch)
tq_ldr = tqdm(self.train_loader, position=index)
2019-08-23 13:42:47 +00:00
_t = time()
for train_data in tq_ldr:
yield self.model
self.do_step(train_data)
2019-08-23 13:42:47 +00:00
if __name__ == '__main__':
2020-10-22 19:27:32 +00:00
parser = argparse.ArgumentParser()
2022-06-10 15:24:41 +00:00
parser.add_argument('-opt', type=str, help='Path to option YAML file.', default='../options/train_music_diffusion_tfd.yml')
2020-10-22 19:27:32 +00:00
parser.add_argument('--launcher', choices=['none', 'pytorch'], default='none', help='job launcher')
args = parser.parse_args()
opt = option.parse(args.opt, is_train=True)
if args.launcher != 'none':
# export CUDA_VISIBLE_DEVICES for running in distributed mode.
if 'gpu_ids' in opt.keys():
gpu_list = ','.join(str(x) for x in opt['gpu_ids'])
os.environ['CUDA_VISIBLE_DEVICES'] = gpu_list
print('export CUDA_VISIBLE_DEVICES=' + gpu_list)
trainer = Trainer()
2021-01-01 18:59:00 +00:00
#### distributed training settings
if args.launcher == 'none': # disabled distributed training
opt['dist'] = False
trainer.rank = -1
if len(opt['gpu_ids']) == 1:
torch.cuda.set_device(opt['gpu_ids'][0])
print('Disabled distributed training.')
else:
opt['dist'] = True
2021-01-01 18:59:00 +00:00
init_dist('nccl')
trainer.world_size = torch.distributed.get_world_size()
trainer.rank = torch.distributed.get_rank()
2022-03-04 19:46:41 +00:00
torch.cuda.set_device(torch.distributed.get_rank())
trainer.init(args.opt, opt, args.launcher)
trainer.do_training()