def test(): model.eval() test_loss = 0. test_accuracy = 0. for data, target in test_loader: if args.cuda: data, target = data.cuda(), target.cuda() output = model(data) # sum up batch loss test_loss += F.nll_loss(output, target, size_average=False).item() # get the index of the max log-probability pred = output.data.max(1, keepdim=True)[1] test_accuracy += pred.eq(target.data.view_as(pred)).cpu().float().sum() # BytePS: use test_sampler to determine the number of examples in # this worker's partition. test_loss /= len(test_sampler) test_accuracy /= len(test_sampler) # BytePS: average metric values across workers. test_loss = metric_average(test_loss, 'avg_loss') test_accuracy = metric_average(test_accuracy, 'avg_accuracy') # BytePS: print output only on first rank. if bps.rank() == 0: print('\nTest set: Average loss: {:.4f}, Accuracy: {:.2f}%\n'.format( test_loss, 100. * test_accuracy))
def save_checkpoint(epoch): if bps.rank() == 0: filepath = args.checkpoint_format.format(epoch=epoch + 1) state = { 'model': model.state_dict(), 'optimizer': optimizer.state_dict(), } torch.save(state, filepath)
def benchmark(tensor, average, name): if not args.no_wait and bps.rank() == 0: time.sleep(0.01) start = time.time() handle = push_pull_async_inplace(tensor, average, name) while True: if poll(handle): synchronize(handle) break end = time.time() return (end - start) * 1000
def save(self, dir, step): if bps.rank() != 0: return params = {} params['genA2B'] = self.genA2B.state_dict() params['genB2A'] = self.genB2A.state_dict() params['disGA'] = self.disGA.state_dict() params['disGB'] = self.disGB.state_dict() params['disLA'] = self.disLA.state_dict() params['disLB'] = self.disLB.state_dict() torch.save(params, os.path.join(dir, self.dataset + '_params_%07d.pt' % step))
def benchmark(tensor, average, name): if not args.no_wait and hvd.rank() == 0: # let other workers submit allreduce request first time.sleep(0.01) start = time.time() # do not use allreduce_() as it polls every 1ms handle = push_pull_async_inplace(tensor, average, name) while True: if poll(handle): synchronize(handle) break end = time.time() return (end - start) * 1000
def check_args(args): # --result_dir if bps.rank() == 0: check_folder(os.path.join(args.result_dir, args.dataset, 'model')) check_folder(os.path.join(args.result_dir, args.dataset, 'img')) check_folder(os.path.join(args.result_dir, args.dataset, 'test')) # --epoch # try: # assert args.epoch >= 1 # except: # print('number of epochs must be larger than or equal to one') # --batch_size try: assert args.batch_size >= 1 except: print('batch size must be larger than or equal to one') return args
cudnn.benchmark = True # If set > 0, will resume training from a given checkpoint. resume_from_epoch = 0 for try_epoch in range(args.epochs, 0, -1): if os.path.exists(args.checkpoint_format.format(epoch=try_epoch)): resume_from_epoch = try_epoch break # BytePS: broadcast resume_from_epoch from rank 0 (which will have # checkpoints) to other ranks. #resume_from_epoch = bps.broadcast(torch.tensor(resume_from_epoch), root_rank=0, # name='resume_from_epoch').item() # BytePS: print logs on the first worker. verbose = 1 if bps.rank() == 0 else 0 # BytePS: write TensorBoard logs on first worker. log_writer = tensorboardX.SummaryWriter(args.log_dir) if bps.rank() == 0 else None kwargs = {'num_workers': 4, 'pin_memory': True} if args.cuda else {} train_dataset = \ datasets.ImageFolder(args.train_dir, transform=transforms.Compose([ transforms.RandomResizedCrop(224), transforms.RandomHorizontalFlip(), transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ]))
def log(s, nl=True): if hvd.rank() != 0: return print(s, end='\n' if nl else '')
def main(): # os.system('shutdown -c') # cancel previous shutdown command log.console(args) tb.log('sizes/world', bps.size()) # need to index validation directory before we start counting the time dataloader.sort_ar(args.data + '/validation') # if args.distributed: # log.console('Distributed initializing process group') torch.cuda.set_device(bps.local_rank()) print(f'cuda device set to {bps.local_rank()}') log.console("cuda initialized (rank=%d)" % (bps.local_rank())) # dist.init_process_group(backend=args.dist_backend, init_method=args.dist_url, world_size=bps.size()) log.console("Distributed: success (%d/%d)" % (bps.rank(), bps.size())) log.console("Loading model (rank=%d)" % (bps.rank())) model = resnet.resnet50(bn0=args.init_bn0).cuda() # reuse the validate tensor global validate_tensor, dist_validate_tensor validate_tensor = torch.tensor([0, 0, 0, 0]).float().cuda() dist_validate_tensor = torch.tensor([0, 0, 0, 0, 0]).float().cuda() if args.fp16: model = network_to_half(model) best_top5 = 93 # only save models over 93%. Otherwise it stops to save every time global model_params, master_params if args.fp16: model_params, master_params = prep_param_lists(model) else: model_params = master_params = model.parameters() optim_params, name_list = experimental_utils.bnwd_optim_params( model, model_params, master_params) if args.no_bn_wd else master_params # define loss function (criterion) and optimizer criterion = nn.CrossEntropyLoss().cuda() optimizer = torch.optim.SGD( optim_params, 0, momentum=args.momentum, weight_decay=args.weight_decay ) # start with 0 lr. Scheduler will change this later named_param = [] for p in optim_params: tensors = p['params'] for tensor in tensors: named_param.append(tensor) # create bps_param (tuple) bps_param = [] for i, tensor in enumerate(named_param): name = name_list[i] bps_param.append((name, tensor)) # wrap with byteps optimizer optimizer = DistributedOptimizer( optimizer, named_parameters=bps_param, backward_passes_per_step=args.batches_per_pushpull, half=True, model=model, fp16_params=model_params, fp32_params=master_params, loss_scale=args.loss_scale) if args.resume: checkpoint = torch.load( args.resume, map_location=lambda storage, loc: storage.cuda(args.local_rank)) model.load_state_dict(checkpoint['state_dict']) args.start_epoch = checkpoint['epoch'] best_top5 = checkpoint['best_top5'] optimizer.load_state_dict(checkpoint['optimizer']) log.console( "Creating data loaders (this could take up to 10 minutes if volume needs to be warmed up)" ) num_machines = (bps.size() - 1) // 8 + 1 assert (num_machines in schedules) phases = schedules[num_machines] dm = DataManager([copy.deepcopy(p) for p in phases if 'bs' in p]) scheduler = Scheduler(optimizer, [copy.deepcopy(p) for p in phases if 'lr' in p]) # BytePS: broadcast parameters & optimizer state. broadcast_parameters([(name, p.detach()) for name, p in bps_param], root_rank=0) broadcast_optimizer_state(optimizer, root_rank=0) start_time = datetime.now() # Loading start to after everything is loaded if args.evaluate: return validate(dm.val_dl, model, criterion, 0, start_time) if args.distributed: log.console('Global Barrier: Syncing machines before training') tensor = torch.tensor([1.0]).float().cuda() barrier_handler = push_pull_async_inplace(tensor, average=True, name="init.barrier") while True: if poll(barrier_handler): synchronize(barrier_handler) break # do broadcast for validate tensor log.console('Broadcasting validate tensor') barrier_handler = push_pull_async_inplace(validate_tensor, average=True, name="validation_tensor") while True: if poll(barrier_handler): synchronize(barrier_handler) break barrier_handler = push_pull_async_inplace( dist_validate_tensor, average=True, name="distributed_validation_tensor") while True: if poll(barrier_handler): synchronize(barrier_handler) break log.event("~~epoch\thours\ttop1\ttop5\n") for epoch in range(args.start_epoch, scheduler.tot_epochs): dm.set_epoch(epoch) train(dm.trn_dl, model, criterion, optimizer, scheduler, epoch) top1, top5 = validate(dm.val_dl, model, criterion, epoch, start_time) time_diff = (datetime.now() - start_time).total_seconds() / 3600.0 log.event(f'~~{epoch}\t{time_diff:.5f}\t\t{top1:.3f}\t\t{top5:.3f}\n') is_best = top5 > best_top5 best_top5 = max(top5, best_top5) if args.local_rank == 0: if is_best: save_checkpoint(epoch, model, best_top5, optimizer, is_best=True, filename='model_best.pth.tar') phase = dm.get_phase(epoch) if phase: save_checkpoint( epoch, model, best_top5, optimizer, filename=f'sz{phase["bs"]}_checkpoint.path.tar')
sys.stdout.flush() log('Model: %s' % args.model) log('Batch size: %d' % args.batch_size) device = 'GPU' if args.cuda else 'CPU' log('Number of %ss: %d' % (device, bps.size())) # Warm-up log('Running warmup...') timeit.timeit(benchmark_step, number=args.num_warmup_batches) # Benchmark log('Running benchmark...') img_secs = [] enable_profiling = args.profiler & (bps.rank() == 0) with torch.autograd.profiler.profile(enable_profiling, True) as prof: for x in range(args.num_iters): time = timeit.timeit(benchmark_step, number=args.num_batches_per_iter) img_sec = args.batch_size * args.num_batches_per_iter / time log('Iter #%d: %.1f img/sec per %s' % (x, img_sec, device)) img_secs.append(img_sec) # Results img_sec_mean = np.mean(img_secs) img_sec_conf = 1.96 * np.std(img_secs) log('Img/sec per %s: %.1f +-%.1f' % (device, img_sec_mean, img_sec_conf)) log('Total img/sec on %d %s(s): %.1f +-%.1f' % (bps.size(), device, bps.size() * img_sec_mean, bps.size() * img_sec_conf))
cudnn.benchmark = True # If set > 0, will resume training from a given checkpoint. resume_from_epoch = 0 for try_epoch in range(args.epochs, 0, -1): if os.path.exists(args.checkpoint_format.format(epoch=try_epoch)): resume_from_epoch = try_epoch break # BytePS: broadcast resume_from_epoch from rank 0 (which will have # checkpoints) to other ranks. #resume_from_epoch = bps.broadcast(torch.tensor(resume_from_epoch), root_rank=0, # name='resume_from_epoch').item() # BytePS: print logs on the first worker. verbose = 1 if bps.rank() == 0 else 0 # BytePS: write TensorBoard logs on first worker. log_writer = tensorboardX.SummaryWriter( args.log_dir) if bps.rank() == 0 else None kwargs = {'num_workers': 4, 'pin_memory': True} if args.cuda else {} train_dataset = \ datasets.ImageFolder(args.train_dir, transform=transforms.Compose([ transforms.RandomResizedCrop(224), transforms.RandomHorizontalFlip(), transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ]))
def main(): parser = argparse.ArgumentParser() # Required parameters parser.add_argument("--train_data_file", default=None, type=str, required=True, help="The input training data file (a text file).") parser.add_argument( "--output_dir", type=str, required=True, help= "The output directory where the model predictions and checkpoints will be written.", ) parser.add_argument( "--model_type", type=str, required=True, help="The model architecture to be trained or fine-tuned.", ) # Other parameters parser.add_argument( "--eval_data_file", default=None, type=str, help= "An optional input evaluation data file to evaluate the perplexity on (a text file).", ) parser.add_argument( "--line_by_line", action="store_true", help= "Whether distinct lines of text in the dataset are to be handled as distinct sequences.", ) parser.add_argument( "--should_continue", action="store_true", help="Whether to continue from latest checkpoint in output_dir") parser.add_argument( "--model_name_or_path", default=None, type=str, help= "The model checkpoint for weights initialization. Leave None if you want to train a model from scratch.", ) parser.add_argument( "--mlm", action="store_true", help= "Train with masked-language modeling loss instead of language modeling." ) parser.add_argument( "--mlm_probability", type=float, default=0.15, help="Ratio of tokens to mask for masked language modeling loss") parser.add_argument( "--config_name", default=None, type=str, help= "Optional pretrained config name or path if not the same as model_name_or_path. If both are None, initialize a new config.", ) parser.add_argument( "--tokenizer_name", default=None, type=str, help= "Optional pretrained tokenizer name or path if not the same as model_name_or_path. If both are None, initialize a new tokenizer.", ) parser.add_argument( "--cache_dir", default=None, type=str, help= "Optional directory to store the pre-trained models downloaded from s3 (instead of the default one)", ) parser.add_argument( "--block_size", default=-1, type=int, help="Optional input sequence length after tokenization." "The training dataset will be truncated in block of this size for training." "Default to the model max input length for single sentence inputs (take into account special tokens).", ) parser.add_argument("--do_train", action="store_true", help="Whether to run training.") parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.") parser.add_argument("--per_gpu_train_batch_size", default=4, type=int, help="Batch size per GPU/CPU for training.") parser.add_argument("--per_gpu_eval_batch_size", default=4, type=int, help="Batch size per GPU/CPU for evaluation.") parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help= "Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument("--num_train_epochs", default=1.0, type=float, help="Total number of training epochs to perform.") parser.add_argument( "--max_steps", default=-1, type=int, help= "If > 0: set total number of training steps to perform. Override num_train_epochs.", ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--logging_steps", type=int, default=500, help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.") parser.add_argument( "--save_total_limit", type=int, default=None, help= "Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default", ) parser.add_argument( "--eval_all_checkpoints", action="store_true", help= "Evaluate all checkpoints starting with the same prefix as model_name_or_path ending and ending with step number", ) parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available") parser.add_argument("--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory") parser.add_argument( "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets") parser.add_argument("--seed", type=int, default=42, help="random seed for initialization") parser.add_argument( "--fp16", action="store_true", help= "Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit", ) parser.add_argument( "--fp16_opt_level", type=str, default="O1", help= "For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html", ) parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.") parser.add_argument("--server_port", type=str, default="", help="For distant debugging.") args = parser.parse_args() # assign local rank args.local_rank = bps.local_rank() if args.model_type in ["bert", "roberta", "distilbert", "camembert" ] and not args.mlm: raise ValueError( "BERT and RoBERTa-like models do not have LM heads but masked LM heads. They must be run using the --mlm " "flag (masked language modeling).") if args.eval_data_file is None and args.do_eval: raise ValueError( "Cannot do evaluation without an evaluation data file. Either supply a file to --eval_data_file " "or remove the --do_eval argument.") if args.should_continue: sorted_checkpoints = _sorted_checkpoints(args) if len(sorted_checkpoints) == 0: raise ValueError( "Used --should_continue but no checkpoint was found in --output_dir." ) else: args.model_name_or_path = sorted_checkpoints[-1] if (os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir): raise ValueError( "Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome." .format(args.output_dir)) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) args.n_gpu = 1 args.device = device # Setup logging logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN, ) logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16, ) # Set seed set_seed(args) # Load pretrained model and tokenizer config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] if args.config_name: config = config_class.from_pretrained(args.config_name, cache_dir=args.cache_dir) elif args.model_name_or_path: config = config_class.from_pretrained(args.model_name_or_path, cache_dir=args.cache_dir) else: config = config_class() if args.tokenizer_name: tokenizer = tokenizer_class.from_pretrained(args.tokenizer_name, cache_dir=args.cache_dir) elif args.model_name_or_path: tokenizer = tokenizer_class.from_pretrained(args.model_name_or_path, cache_dir=args.cache_dir) else: raise ValueError( "You are instantiating a new {} tokenizer. This is not supported, but you can do it from another script, save it," "and load it from here, using --tokenizer_name".format( tokenizer_class.__name__)) if args.block_size <= 0: args.block_size = tokenizer.max_len_single_sentence # Our input block size will be the max possible for the model else: args.block_size = min(args.block_size, tokenizer.max_len_single_sentence) if args.model_name_or_path: model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool(".ckpt" in args.model_name_or_path), config=config, cache_dir=args.cache_dir, ) else: logger.info("Training new model from scratch") model = model_class(config=config) model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Training if args.do_train: train_dataset = load_and_cache_examples(args, tokenizer, evaluate=False) global_step, tr_loss = train(args, train_dataset, model, tokenizer) logger.info(" global_step = %s, average loss = %s", global_step, tr_loss) # Saving best-practices: if you use save_pretrained for the model and tokenizer, you can reload them using from_pretrained() if args.do_train and (args.local_rank == -1 or bps.rank() == 0): # Create output directory if needed if args.local_rank in [-1, 0]: os.makedirs(args.output_dir, exist_ok=True) logger.info("Saving model checkpoint to %s", args.output_dir) # Save a trained model, configuration and tokenizer using `save_pretrained()`. # They can then be reloaded using `from_pretrained()` model_to_save = (model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(args.output_dir) tokenizer.save_pretrained(args.output_dir) # Good practice: save your training arguments together with the trained model torch.save(args, os.path.join(args.output_dir, "training_args.bin")) # Load a trained model and vocabulary that you have fine-tuned model = model_class.from_pretrained(args.output_dir) tokenizer = tokenizer_class.from_pretrained(args.output_dir) model.to(args.device) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted( glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True))) logging.getLogger("transformers.modeling_utils").setLevel( logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split( "-")[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split( "/")[-1] if checkpoint.find("checkpoint") != -1 else "" model = model_class.from_pretrained(checkpoint) model.to(args.device) result = evaluate(args, model, tokenizer, prefix=prefix) result = dict( (k + "_{}".format(global_step), v) for k, v in result.items()) results.update(result) return results
def main_worker(gpu, ngpus_per_node, args): global best_acc1 args.gpu = gpu if args.gpu is not None: print("Use GPU: {} for training".format(args.gpu)) if args.distributed: if args.dist_url == "env://" and args.rank == -1: args.rank = int(os.environ["RANK"]) if args.multiprocessing_distributed: # For multiprocessing distributed training, rank needs to be the # global rank among all the processes args.rank = args.rank * ngpus_per_node + gpu # create model if args.pretrained: print("=> using pre-trained model '{}'".format(args.arch)) model = models.__dict__[args.arch](pretrained=True) else: print("=> creating model '{}'".format(args.arch)) model = models.__dict__[args.arch]() if args.distributed: # For multiprocessing distributed, DistributedDataParallel constructor # should always set the single device scope, otherwise, # DistributedDataParallel will use all available devices. if args.gpu is not None: torch.cuda.set_device(args.gpu) model.cuda(args.gpu) # When using a single GPU per process and per # DistributedDataParallel, we need to divide the batch size # ourselves based on the total number of GPUs we have args.batch_size = int(args.batch_size / ngpus_per_node) args.workers = int( (args.workers + ngpus_per_node - 1) / ngpus_per_node) model = DDP(model, device_ids=[args.gpu], broadcast_buffers=args.broadcast_buffers) # define loss function (criterion) and optimizer criterion = nn.CrossEntropyLoss().cuda(args.gpu) optimizer = torch.optim.SGD(model.parameters(), args.lr, momentum=args.momentum, weight_decay=args.weight_decay) # optionally resume from a checkpoint if args.resume: if os.path.isfile(args.resume): print("=> loading checkpoint '{}'".format(args.resume)) if args.gpu is None: checkpoint = torch.load(args.resume) else: # Map model to be loaded to specified single gpu. loc = 'cuda:{}'.format(args.gpu) checkpoint = torch.load(args.resume, map_location=loc) args.start_epoch = checkpoint['epoch'] best_acc1 = checkpoint['best_acc1'] if args.gpu is not None: # best_acc1 may be from a checkpoint from a different GPU best_acc1 = best_acc1.to(args.gpu) model.load_state_dict(checkpoint['state_dict']) optimizer.load_state_dict(checkpoint['optimizer']) print("=> loaded checkpoint '{}' (epoch {})".format( args.resume, checkpoint['epoch'])) else: print("=> no checkpoint found at '{}'".format(args.resume)) cudnn.benchmark = True # Data loading code traindir = os.path.join(args.data, 'train') valdir = os.path.join(args.data, 'val') normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) train_dataset = datasets.ImageFolder( traindir, transforms.Compose([ transforms.RandomResizedCrop(224), transforms.RandomHorizontalFlip(), transforms.ToTensor(), normalize, ])) if args.distributed: train_sampler = torch.utils.data.distributed.DistributedSampler( train_dataset, num_replicas=bps.size(), rank=bps.rank()) else: train_sampler = None train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, shuffle=(train_sampler is None), num_workers=args.workers, pin_memory=True, sampler=train_sampler) val_loader = torch.utils.data.DataLoader(datasets.ImageFolder( valdir, transforms.Compose([ transforms.Resize(256), transforms.CenterCrop(224), transforms.ToTensor(), normalize, ])), batch_size=args.batch_size, shuffle=False, num_workers=args.workers, pin_memory=True) if args.evaluate: validate(val_loader, model, criterion, args) return for epoch in range(args.start_epoch, args.epochs): if args.distributed: train_sampler.set_epoch(epoch) adjust_learning_rate(optimizer, epoch, args) # train for one epoch train(train_loader, model, criterion, optimizer, epoch, args) # evaluate on validation set acc1 = validate(val_loader, model, criterion, args) # remember best acc@1 and save checkpoint is_best = acc1 > best_acc1 best_acc1 = max(acc1, best_acc1) if not args.multiprocessing_distributed or ( args.multiprocessing_distributed and args.rank % ngpus_per_node == 0): save_checkpoint( { 'epoch': epoch + 1, 'arch': args.arch, 'state_dict': model.state_dict(), 'best_acc1': best_acc1, 'optimizer': optimizer.state_dict(), }, is_best)
schedules = { 1: one_machine, 2: two_machine, 4: four_machines, 8: eight_machines, 16: sixteen_machines } bps.init() cudnn.benchmark = True args = get_parser().parse_args() # Only want master rank logging to tensorboard is_master = (not args.distributed) or (bps.rank() == 0) is_rank0 = bps.local_rank() == 0 tb = TensorboardLogger(args.logdir, is_master=is_master) log = FileLogger(args.logdir, is_master=is_master, is_rank0=is_rank0) def main(): # os.system('shutdown -c') # cancel previous shutdown command log.console(args) tb.log('sizes/world', bps.size()) # need to index validation directory before we start counting the time dataloader.sort_ar(args.data + '/validation') # if args.distributed: # log.console('Distributed initializing process group')
def train(args, train_dataset, model: PreTrainedModel, tokenizer: PreTrainedTokenizer) -> Tuple[int, float]: """ Train the model """ if args.local_rank in [-1, 0]: tb_writer = SummaryWriter() args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu) def collate(examples: List[torch.Tensor]): if tokenizer._pad_token is None: return pad_sequence(examples, batch_first=True) return pad_sequence(examples, batch_first=True, padding_value=tokenizer.pad_token_id) train_sampler = DistributedSampler(train_dataset, num_replicas=bps.size(), rank=bps.rank()) train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size, collate_fn=collate) if args.max_steps > 0: t_total = args.max_steps args.num_train_epochs = args.max_steps // ( len(train_dataloader) // args.gradient_accumulation_steps) + 1 else: t_total = len( train_dataloader ) // args.gradient_accumulation_steps * args.num_train_epochs # Prepare optimizer and schedule (linear warmup and decay) no_decay = ["bias", "LayerNorm.weight"] optimizer_grouped_parameters = [ { "params": [ p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay) ], "weight_decay": args.weight_decay, }, { "params": [ p for n, p in model.named_parameters() if any(nd in n for nd in no_decay) ], "weight_decay": 0.0 }, ] optimizer = SGD(optimizer_grouped_parameters, lr=args.learning_rate, momentum=0.9) optimizer = bps.DistributedOptimizer( optimizer, named_parameters=model.named_parameters()) bps.broadcast_parameters(model.state_dict(), root_rank=0) scheduler = get_linear_schedule_with_warmup( optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total) # Check if saved optimizer or scheduler states exist if (args.model_name_or_path and os.path.isfile( os.path.join(args.model_name_or_path, "optimizer.pt")) and os.path.isfile( os.path.join(args.model_name_or_path, "scheduler.pt"))): # Load in optimizer and scheduler states optimizer.load_state_dict( torch.load(os.path.join(args.model_name_or_path, "optimizer.pt"))) scheduler.load_state_dict( torch.load(os.path.join(args.model_name_or_path, "scheduler.pt"))) if args.fp16: try: from apex import amp except ImportError: raise ImportError( "Please install apex from https://www.github.com/nvidia/apex to use fp16 training." ) model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level) # Train! logger.info("***** Running training *****") logger.info(" Num examples = %d", len(train_dataset)) logger.info(" Num Epochs = %d", args.num_train_epochs) logger.info(" Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size) logger.info( " Total train batch size (w. parallel, distributed & accumulation) = %d", args.train_batch_size * args.gradient_accumulation_steps * (bps.size() if args.local_rank != -1 else 1), ) logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps) logger.info(" Total optimization steps = %d", t_total) global_step = 0 epochs_trained = 0 steps_trained_in_current_epoch = 0 # Check if continuing training from a checkpoint if args.model_name_or_path and os.path.exists(args.model_name_or_path): try: # set global_step to gobal_step of last saved checkpoint from model path checkpoint_suffix = args.model_name_or_path.split("-")[-1].split( "/")[0] global_step = int(checkpoint_suffix) epochs_trained = global_step // (len(train_dataloader) // args.gradient_accumulation_steps) steps_trained_in_current_epoch = global_step % ( len(train_dataloader) // args.gradient_accumulation_steps) logger.info( " Continuing training from checkpoint, will skip to saved global_step" ) logger.info(" Continuing training from epoch %d", epochs_trained) logger.info(" Continuing training from global step %d", global_step) logger.info(" Will skip the first %d steps in the first epoch", steps_trained_in_current_epoch) except ValueError: logger.info(" Starting fine-tuning.") tr_loss, logging_loss = 0.0, 0.0 model_to_resize = model.module if hasattr( model, "module") else model # Take care of distributed/parallel training model_to_resize.resize_token_embeddings(len(tokenizer)) model.zero_grad() train_iterator = trange(epochs_trained, int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0]) set_seed(args) # Added here for reproducibility for _ in train_iterator: epoch_iterator = tqdm(train_dataloader, desc="Iteration", disable=args.local_rank not in [-1, 0]) for step, batch in enumerate(epoch_iterator): # Skip past any already trained steps if resuming training if steps_trained_in_current_epoch > 0: steps_trained_in_current_epoch -= 1 continue inputs, labels = mask_tokens(batch, tokenizer, args) if args.mlm else (batch, batch) inputs = inputs.to(args.device) labels = labels.to(args.device) model.train() outputs = model(inputs, masked_lm_labels=labels) if args.mlm else model( inputs, labels=labels) loss = outputs[ 0] # model outputs are always tuple in transformers (see doc) if args.n_gpu > 1: loss = loss.mean( ) # mean() to average on multi-gpu parallel training if args.gradient_accumulation_steps > 1: loss = loss / args.gradient_accumulation_steps if args.fp16: with amp.scale_loss(loss, optimizer) as scaled_loss: scaled_loss.backward() else: loss.backward() tr_loss += loss.item() if (step + 1) % args.gradient_accumulation_steps == 0: if args.fp16: torch.nn.utils.clip_grad_norm_( amp.master_params(optimizer), args.max_grad_norm) else: torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm) optimizer.step() scheduler.step() # Update learning rate schedule model.zero_grad() global_step += 1 if args.local_rank in [ -1, 0 ] and args.logging_steps > 0 and global_step % args.logging_steps == 0: # Log metrics if ( args.local_rank == -1 and args.evaluate_during_training ): # Only evaluate when single GPU otherwise metrics may not average well results = evaluate(args, model, tokenizer) for key, value in results.items(): tb_writer.add_scalar("eval_{}".format(key), value, global_step) tb_writer.add_scalar("lr", scheduler.get_lr()[0], global_step) tb_writer.add_scalar("loss", (tr_loss - logging_loss) / args.logging_steps, global_step) logging_loss = tr_loss if args.local_rank in [ -1, 0 ] and args.save_steps > 0 and global_step % args.save_steps == 0: checkpoint_prefix = "checkpoint" # Save model checkpoint output_dir = os.path.join( args.output_dir, "{}-{}".format(checkpoint_prefix, global_step)) os.makedirs(output_dir, exist_ok=True) model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(output_dir) tokenizer.save_pretrained(output_dir) torch.save(args, os.path.join(output_dir, "training_args.bin")) logger.info("Saving model checkpoint to %s", output_dir) _rotate_checkpoints(args, checkpoint_prefix) torch.save(optimizer.state_dict(), os.path.join(output_dir, "optimizer.pt")) torch.save(scheduler.state_dict(), os.path.join(output_dir, "scheduler.pt")) logger.info("Saving optimizer and scheduler states to %s", output_dir) if args.max_steps > 0 and global_step > args.max_steps: epoch_iterator.close() break if args.max_steps > 0 and global_step > args.max_steps: train_iterator.close() break if args.local_rank in [-1, 0]: tb_writer.close() return global_step, tr_loss / global_step
help='use fp16 compression during pushpull') args = parser.parse_args() args.cuda = not args.no_cuda and torch.cuda.is_available() # BytePS: initialize library. bps.init() torch.manual_seed(args.seed) if args.cuda: # BytePS: pin GPU to local rank. torch.cuda.set_device(bps.local_rank()) torch.cuda.manual_seed(args.seed) kwargs = {'num_workers': 1, 'pin_memory': True} if args.cuda else {} train_dataset = \ datasets.MNIST('data-%d' % bps.rank(), train=True, download=True, transform=transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,)) ])) # BytePS: use DistributedSampler to partition the training data. train_sampler = torch.utils.data.distributed.DistributedSampler( train_dataset, num_replicas=bps.size(), rank=bps.rank()) train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, sampler=train_sampler, **kwargs) test_dataset = \ datasets.MNIST('data-%d' % bps.rank(), train=False, transform=transforms.Compose([ transforms.ToTensor(),
def train(self): self.genA2B.train(), self.genB2A.train(), self.disGA.train( ), self.disGB.train(), self.disLA.train(), self.disLB.train() start_iter = 1 if self.resume: model_list = glob( os.path.join(self.result_dir, self.dataset, 'model', '*.pt')) if not len(model_list) == 0: model_list.sort() start_iter = int(model_list[-1].split('_')[-1].split('.')[0]) self.load(os.path.join(self.result_dir, self.dataset, 'model'), start_iter) print(" [*] Load SUCCESS") if self.decay_flag and start_iter > (self.iteration // 2): self.G_optim.param_groups[0]['lr'] -= ( self.lr / (self.iteration // 2)) * (start_iter - self.iteration // 2) self.D_optim.param_groups[0]['lr'] -= ( self.lr / (self.iteration // 2)) * (start_iter - self.iteration // 2) # training loop print('training start !') start_time = time.time() last_time = start_time for step in range(start_iter, self.iteration + 1): if self.decay_flag and step > (self.iteration // 2): self.G_optim.param_groups[0]['lr'] -= (self.lr / (self.iteration // 2)) self.D_optim.param_groups[0]['lr'] -= (self.lr / (self.iteration // 2)) try: real_A, _ = trainA_iter.next() except: trainA_iter = iter(self.trainA_loader) real_A, _ = trainA_iter.next() try: real_B, _ = trainB_iter.next() except: trainB_iter = iter(self.trainB_loader) real_B, _ = trainB_iter.next() real_A, real_B = real_A.to(self.device), real_B.to(self.device) # Update D self.D_optim._handles.clear() self.D_optim.zero_grad() self.D_optim.set_backward_passes_per_step(1) self.G_optim.set_backward_passes_per_step(10) fake_A2B, _, _ = self.genA2B(real_A) fake_B2A, _, _ = self.genB2A(real_B) real_GA_logit, real_GA_cam_logit, _ = self.disGA(real_A) real_LA_logit, real_LA_cam_logit, _ = self.disLA(real_A) real_GB_logit, real_GB_cam_logit, _ = self.disGB(real_B) real_LB_logit, real_LB_cam_logit, _ = self.disLB(real_B) fake_GA_logit, fake_GA_cam_logit, _ = self.disGA(fake_B2A) fake_LA_logit, fake_LA_cam_logit, _ = self.disLA(fake_B2A) fake_GB_logit, fake_GB_cam_logit, _ = self.disGB(fake_A2B) fake_LB_logit, fake_LB_cam_logit, _ = self.disLB(fake_A2B) D_ad_loss_GA = self.MSE_loss( real_GA_logit, torch.ones_like(real_GA_logit).to( self.device)) + self.MSE_loss( fake_GA_logit, torch.zeros_like(fake_GA_logit).to(self.device)) D_ad_cam_loss_GA = self.MSE_loss( real_GA_cam_logit, torch.ones_like(real_GA_cam_logit).to( self.device)) + self.MSE_loss( fake_GA_cam_logit, torch.zeros_like(fake_GA_cam_logit).to(self.device)) D_ad_loss_LA = self.MSE_loss( real_LA_logit, torch.ones_like(real_LA_logit).to( self.device)) + self.MSE_loss( fake_LA_logit, torch.zeros_like(fake_LA_logit).to(self.device)) D_ad_cam_loss_LA = self.MSE_loss( real_LA_cam_logit, torch.ones_like(real_LA_cam_logit).to( self.device)) + self.MSE_loss( fake_LA_cam_logit, torch.zeros_like(fake_LA_cam_logit).to(self.device)) D_ad_loss_GB = self.MSE_loss( real_GB_logit, torch.ones_like(real_GB_logit).to( self.device)) + self.MSE_loss( fake_GB_logit, torch.zeros_like(fake_GB_logit).to(self.device)) D_ad_cam_loss_GB = self.MSE_loss( real_GB_cam_logit, torch.ones_like(real_GB_cam_logit).to( self.device)) + self.MSE_loss( fake_GB_cam_logit, torch.zeros_like(fake_GB_cam_logit).to(self.device)) D_ad_loss_LB = self.MSE_loss( real_LB_logit, torch.ones_like(real_LB_logit).to( self.device)) + self.MSE_loss( fake_LB_logit, torch.zeros_like(fake_LB_logit).to(self.device)) D_ad_cam_loss_LB = self.MSE_loss( real_LB_cam_logit, torch.ones_like(real_LB_cam_logit).to( self.device)) + self.MSE_loss( fake_LB_cam_logit, torch.zeros_like(fake_LB_cam_logit).to(self.device)) D_loss_A = self.adv_weight * (D_ad_loss_GA + D_ad_cam_loss_GA + D_ad_loss_LA + D_ad_cam_loss_LA) D_loss_B = self.adv_weight * (D_ad_loss_GB + D_ad_cam_loss_GB + D_ad_loss_LB + D_ad_cam_loss_LB) Discriminator_loss = D_loss_A + D_loss_B Discriminator_loss.backward() self.D_optim.step() # Update G self.G_optim._handles.clear() self.G_optim.zero_grad() self.D_optim.set_backward_passes_per_step(10) self.G_optim.set_backward_passes_per_step(1) fake_A2B, fake_A2B_cam_logit, _ = self.genA2B(real_A) fake_B2A, fake_B2A_cam_logit, _ = self.genB2A(real_B) fake_A2B2A, _, _ = self.genB2A(fake_A2B) fake_B2A2B, _, _ = self.genA2B(fake_B2A) fake_A2A, fake_A2A_cam_logit, _ = self.genB2A(real_A) fake_B2B, fake_B2B_cam_logit, _ = self.genA2B(real_B) fake_GA_logit, fake_GA_cam_logit, _ = self.disGA(fake_B2A) fake_LA_logit, fake_LA_cam_logit, _ = self.disLA(fake_B2A) fake_GB_logit, fake_GB_cam_logit, _ = self.disGB(fake_A2B) fake_LB_logit, fake_LB_cam_logit, _ = self.disLB(fake_A2B) G_ad_loss_GA = self.MSE_loss( fake_GA_logit, torch.ones_like(fake_GA_logit).to(self.device)) G_ad_cam_loss_GA = self.MSE_loss( fake_GA_cam_logit, torch.ones_like(fake_GA_cam_logit).to(self.device)) G_ad_loss_LA = self.MSE_loss( fake_LA_logit, torch.ones_like(fake_LA_logit).to(self.device)) G_ad_cam_loss_LA = self.MSE_loss( fake_LA_cam_logit, torch.ones_like(fake_LA_cam_logit).to(self.device)) G_ad_loss_GB = self.MSE_loss( fake_GB_logit, torch.ones_like(fake_GB_logit).to(self.device)) G_ad_cam_loss_GB = self.MSE_loss( fake_GB_cam_logit, torch.ones_like(fake_GB_cam_logit).to(self.device)) G_ad_loss_LB = self.MSE_loss( fake_LB_logit, torch.ones_like(fake_LB_logit).to(self.device)) G_ad_cam_loss_LB = self.MSE_loss( fake_LB_cam_logit, torch.ones_like(fake_LB_cam_logit).to(self.device)) G_recon_loss_A = self.L1_loss(fake_A2B2A, real_A) G_recon_loss_B = self.L1_loss(fake_B2A2B, real_B) G_identity_loss_A = self.L1_loss(fake_A2A, real_A) G_identity_loss_B = self.L1_loss(fake_B2B, real_B) G_cam_loss_A = self.BCE_loss( fake_B2A_cam_logit, torch.ones_like(fake_B2A_cam_logit).to( self.device)) + self.BCE_loss( fake_A2A_cam_logit, torch.zeros_like(fake_A2A_cam_logit).to(self.device)) G_cam_loss_B = self.BCE_loss( fake_A2B_cam_logit, torch.ones_like(fake_A2B_cam_logit).to( self.device)) + self.BCE_loss( fake_B2B_cam_logit, torch.zeros_like(fake_B2B_cam_logit).to(self.device)) G_loss_A = self.adv_weight * ( G_ad_loss_GA + G_ad_cam_loss_GA + G_ad_loss_LA + G_ad_cam_loss_LA ) + self.cycle_weight * G_recon_loss_A + self.identity_weight * G_identity_loss_A + self.cam_weight * G_cam_loss_A G_loss_B = self.adv_weight * ( G_ad_loss_GB + G_ad_cam_loss_GB + G_ad_loss_LB + G_ad_cam_loss_LB ) + self.cycle_weight * G_recon_loss_B + self.identity_weight * G_identity_loss_B + self.cam_weight * G_cam_loss_B Generator_loss = G_loss_A + G_loss_B Generator_loss.backward() self.G_optim.step() # clip parameter of AdaILN and ILN, applied after optimizer step self.genA2B.apply(self.Rho_clipper) self.genB2A.apply(self.Rho_clipper) if bps.local_rank() == 0: this_time = time.time() print( "[%5d/%5d] time: %4.4f speed: %4.4f (sec/step) d_loss: %.8f, g_loss: %.8f" % (step, self.iteration, this_time - start_time, this_time - last_time, Discriminator_loss, Generator_loss)) last_time = this_time if step % self.print_freq == 0: train_sample_num = 5 test_sample_num = 5 A2B = np.zeros((self.img_size * 7, 0, 3)) B2A = np.zeros((self.img_size * 7, 0, 3)) self.genA2B.eval(), self.genB2A.eval(), self.disGA.eval( ), self.disGB.eval(), self.disLA.eval(), self.disLB.eval() for _ in range(train_sample_num): try: real_A, _ = trainA_iter.next() except: trainA_iter = iter(self.trainA_loader) real_A, _ = trainA_iter.next() try: real_B, _ = trainB_iter.next() except: trainB_iter = iter(self.trainB_loader) real_B, _ = trainB_iter.next() real_A, real_B = real_A.to(self.device), real_B.to( self.device) fake_A2B, _, fake_A2B_heatmap = self.genA2B(real_A) fake_B2A, _, fake_B2A_heatmap = self.genB2A(real_B) fake_A2B2A, _, fake_A2B2A_heatmap = self.genB2A(fake_A2B) fake_B2A2B, _, fake_B2A2B_heatmap = self.genA2B(fake_B2A) fake_A2A, _, fake_A2A_heatmap = self.genB2A(real_A) fake_B2B, _, fake_B2B_heatmap = self.genA2B(real_B) A2B = np.concatenate( (A2B, np.concatenate( (RGB2BGR(tensor2numpy(denorm(real_A[0]))), cam(tensor2numpy(fake_A2A_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_A2A[0]))), cam(tensor2numpy(fake_A2B_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_A2B[0]))), cam(tensor2numpy(fake_A2B2A_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_A2B2A[0])))), 0)), 1) B2A = np.concatenate( (B2A, np.concatenate( (RGB2BGR(tensor2numpy(denorm(real_B[0]))), cam(tensor2numpy(fake_B2B_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_B2B[0]))), cam(tensor2numpy(fake_B2A_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_B2A[0]))), cam(tensor2numpy(fake_B2A2B_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_B2A2B[0])))), 0)), 1) for _ in range(test_sample_num): try: real_A, _ = testA_iter.next() except: testA_iter = iter(self.testA_loader) real_A, _ = testA_iter.next() try: real_B, _ = testB_iter.next() except: testB_iter = iter(self.testB_loader) real_B, _ = testB_iter.next() real_A, real_B = real_A.to(self.device), real_B.to( self.device) fake_A2B, _, fake_A2B_heatmap = self.genA2B(real_A) fake_B2A, _, fake_B2A_heatmap = self.genB2A(real_B) fake_A2B2A, _, fake_A2B2A_heatmap = self.genB2A(fake_A2B) fake_B2A2B, _, fake_B2A2B_heatmap = self.genA2B(fake_B2A) fake_A2A, _, fake_A2A_heatmap = self.genB2A(real_A) fake_B2B, _, fake_B2B_heatmap = self.genA2B(real_B) A2B = np.concatenate( (A2B, np.concatenate( (RGB2BGR(tensor2numpy(denorm(real_A[0]))), cam(tensor2numpy(fake_A2A_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_A2A[0]))), cam(tensor2numpy(fake_A2B_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_A2B[0]))), cam(tensor2numpy(fake_A2B2A_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_A2B2A[0])))), 0)), 1) B2A = np.concatenate( (B2A, np.concatenate( (RGB2BGR(tensor2numpy(denorm(real_B[0]))), cam(tensor2numpy(fake_B2B_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_B2B[0]))), cam(tensor2numpy(fake_B2A_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_B2A[0]))), cam(tensor2numpy(fake_B2A2B_heatmap[0]), self.img_size), RGB2BGR(tensor2numpy(denorm(fake_B2A2B[0])))), 0)), 1) if bps.rank() == 0: cv2.imwrite( os.path.join(self.result_dir, self.dataset, 'img', 'A2B_%07d.png' % step), A2B * 255.0) cv2.imwrite( os.path.join(self.result_dir, self.dataset, 'img', 'B2A_%07d.png' % step), B2A * 255.0) self.genA2B.train(), self.genB2A.train(), self.disGA.train( ), self.disGB.train(), self.disLA.train(), self.disLB.train() if (step % self.save_freq == 0) and (bps.rank() == 0): self.save(os.path.join(self.result_dir, self.dataset, 'model'), step) if (step % 1000 == 0) and (bps.rank() == 0): params = {} params['genA2B'] = self.genA2B.state_dict() params['genB2A'] = self.genB2A.state_dict() params['disGA'] = self.disGA.state_dict() params['disGB'] = self.disGB.state_dict() params['disLA'] = self.disLA.state_dict() params['disLB'] = self.disLB.state_dict() torch.save( params, os.path.join(self.result_dir, self.dataset + '_params_latest.pt')) if (bps.rank() == 0): self.save(os.path.join(self.result_dir, self.dataset, 'model'), step)
def env_rank(): return bps.rank()
else: if args.cuda: args.model_device = torch.device('cuda') else: args.model_device = torch.device('cpu') # Initialize Horovod/Cuda myrank = 0 mysize = 1 if args.par == "hvd": hvd.init() myrank = hvd.rank() mysize = hvd.size() elif args.par == "bps": bps.init() myrank = bps.rank() mysize = bps.size() torch.manual_seed(args.seed) if args.cuda: # Horovod & BytePS: pin GPU to local rank. if args.par == "hvd": torch.cuda.set_device(hvd.local_rank()) torch.cuda.manual_seed(args.seed) if args.par == "bps": torch.cuda.set_device(bps.local_rank()) torch.cuda.manual_seed(args.seed) # Model definition model = MortgageNetwork( args.num_features, args.embedding_size,
def build_model(self): """ DataLoader """ if self.fix_aug: print("FIX AUG ON") train_transform = transforms.Compose([ transforms.RandomHorizontalFlip(), transforms.Resize((self.img_size, self.img_size)), transforms.ToTensor(), transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)) ]) else: train_transform = transforms.Compose([ transforms.RandomHorizontalFlip(), transforms.Resize((self.img_size + 30, self.img_size + 30)), transforms.RandomCrop(self.img_size), transforms.ToTensor(), transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)) ]) test_transform = transforms.Compose([ transforms.Resize((self.img_size, self.img_size)), transforms.ToTensor(), transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)) ]) self.trainA = ImageFolder(os.path.join(self.dataset_dir, self.dataset, 'trainA'), train_transform, list_mode=self.list_mode) self.trainB = ImageFolder(os.path.join(self.dataset_dir, self.dataset, 'trainB'), train_transform, list_mode=self.list_mode) self.testA = ImageFolder(os.path.join(self.dataset_dir, self.dataset, 'testA'), test_transform, list_mode=self.list_mode) self.testB = ImageFolder(os.path.join(self.dataset_dir, self.dataset, 'testB'), test_transform, list_mode=self.list_mode) trainA_sampler = torch.utils.data.distributed.DistributedSampler( self.trainA, num_replicas=bps.size(), rank=bps.rank()) trainB_sampler = torch.utils.data.distributed.DistributedSampler( self.trainB, num_replicas=bps.size(), rank=bps.rank()) testA_sampler = torch.utils.data.distributed.DistributedSampler( self.testA, num_replicas=bps.size(), rank=bps.rank()) testB_sampler = torch.utils.data.distributed.DistributedSampler( self.testB, num_replicas=bps.size(), rank=bps.rank()) self.trainA_loader = DataLoader(self.trainA, batch_size=self.batch_size, sampler=trainA_sampler, num_workers=1) self.trainB_loader = DataLoader(self.trainB, batch_size=self.batch_size, sampler=trainB_sampler, num_workers=1) self.testA_loader = DataLoader(self.testA, batch_size=1, sampler=testA_sampler) self.testB_loader = DataLoader(self.testB, batch_size=1, sampler=testB_sampler) """ Define Generator, Discriminator """ self.genA2B = ResnetGenerator(input_nc=3, output_nc=3, ngf=self.ch, n_blocks=self.n_res, img_size=self.img_size, light=self.light).to(self.device) self.genB2A = ResnetGenerator(input_nc=3, output_nc=3, ngf=self.ch, n_blocks=self.n_res, img_size=self.img_size, light=self.light).to(self.device) self.disGA = Discriminator(input_nc=3, ndf=self.ch, n_layers=7).to(self.device) self.disGB = Discriminator(input_nc=3, ndf=self.ch, n_layers=7).to(self.device) self.disLA = Discriminator(input_nc=3, ndf=self.ch, n_layers=5).to(self.device) self.disLB = Discriminator(input_nc=3, ndf=self.ch, n_layers=5).to(self.device) """ Define Loss """ self.L1_loss = nn.L1Loss().to(self.device) self.MSE_loss = nn.MSELoss().to(self.device) self.BCE_loss = nn.BCEWithLogitsLoss().to(self.device) gen_named_parameters = [] dis_named_parameters = [] for n, p in (list(self.genA2B.named_parameters(prefix='genA2B')) + list(self.genB2A.named_parameters(prefix='genB2A'))): gen_named_parameters.append((n, p)) for n, p in (list(self.disGA.named_parameters(prefix='disGA')) + list(self.disGB.named_parameters(prefix='disGB')) + list(self.disLA.named_parameters(prefix='disLA')) + list(self.disLB.named_parameters(prefix='disLB'))): dis_named_parameters.append((n, p)) gen_state_dict = OrderedDict( [("genA2B." + k, v) for k, v in self.genA2B.state_dict().items()] + [("genB2A." + k, v) for k, v in self.genB2A.state_dict().items()]) dis_state_dict = OrderedDict( [("disGA." + k, v) for k, v in self.disGA.state_dict().items()] + [("disGB." + k, v) for k, v in self.disGB.state_dict().items()] + [("disLA." + k, v) for k, v in self.disLA.state_dict().items()] + [("disLB." + k, v) for k, v in self.disLB.state_dict().items()]) bps.broadcast_parameters(gen_state_dict, root_rank=0) bps.broadcast_parameters(dis_state_dict, root_rank=0) """ Trainer """ self.G_optim = torch.optim.Adam(itertools.chain( self.genA2B.parameters(), self.genB2A.parameters()), lr=self.lr, betas=(0.5, 0.999), weight_decay=self.weight_decay) self.D_optim = torch.optim.Adam(itertools.chain( self.disGA.parameters(), self.disGB.parameters(), self.disLA.parameters(), self.disLB.parameters()), lr=self.lr, betas=(0.5, 0.999), weight_decay=self.weight_decay) named_parameters = [] for n, p in list(self.genA2B.named_parameters()): named_parameters.append(("genA2B." + n, p)) for n, p in list(self.genB2A.named_parameters()): named_parameters.append(("genB2A." + n, p)) self.G_optim = bps.DistributedOptimizer( self.G_optim, named_parameters=gen_named_parameters, compression=bps.Compression.none) self.D_optim = bps.DistributedOptimizer( self.D_optim, named_parameters=dis_named_parameters, compression=bps.Compression.none) self.G_optim._handles.clear() self.D_optim._handles.clear() """ Define Rho clipper to constraint the value of rho in AdaILN and ILN""" self.Rho_clipper = RhoClipper(0, 1)