def train(args, train_dataset, model, tokenizer): """ 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) train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset) train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size) 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 = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) 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 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) # multi-gpu training (should be after apex fp16 initialization) if args.n_gpu > 1: model = torch.nn.DataParallel(model) # Distributed training (should be after apex fp16 initialization) if args.local_rank != -1: model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True ) # 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 * (torch.distributed.get_world_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 = 1 epochs_trained = 0 steps_trained_in_current_epoch = 0 # Check if continuing training from a checkpoint if 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.zero_grad() train_iterator = trange( epochs_trained, int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0] ) # Added here for reproductibility set_seed(args) 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 model.train() batch = tuple(t.to(args.device) for t in batch) inputs = { "input_ids": batch[0], "attention_mask": batch[1], "token_type_ids": batch[2], "start_positions": batch[3], "end_positions": batch[4], } if args.model_type in ["xlm", "roberta", "distilbert"]: del inputs["token_type_ids"] if args.model_type in ["xlnet", "xlm"]: inputs.update({"cls_index": batch[5], "p_mask": batch[6]}) if args.version_2_with_negative: inputs.update({"is_impossible": batch[7]}) if hasattr(model, "config") and hasattr(model.config, "lang2id"): inputs.update( {"langs": (torch.ones(batch[0].shape, dtype=torch.int64) * args.lang_id).to(args.device)} ) outputs = model(**inputs) # model outputs are always tuple in transformers (see doc) loss = outputs[0] if args.n_gpu > 1: loss = loss.mean() # mean() to average on multi-gpu parallel (not distributed) 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 # Log metrics if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0: # Only evaluate when single GPU otherwise metrics may not average well if args.local_rank == -1 and args.evaluate_during_training: 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 # Save model checkpoint if args.local_rank in [-1, 0] and args.save_steps > 0 and global_step % args.save_steps == 0: output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step)) if not os.path.exists(output_dir): os.makedirs(output_dir) # Take care of distributed/parallel training model_to_save = model.module if hasattr(model, "module") else model 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) 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
def train(args, model, tokenizer): """ Train the model """ train_dataset, train_data_processor = load_examples(args, tokenizer, evaluate="train") labels = train_data_processor.labels label_frequences = get_label_frequencies(train_dataset) label_frequences = [label_frequences[l] for l in labels] label_weights = ( torch.tensor(label_frequences, device=args.device, dtype=torch.float) / len(train_dataset))**-1 criterion = nn.BCEWithLogitsLoss(pos_weight=label_weights) if args.local_rank in [-1, 0]: tb_writer = SummaryWriter() if args.max_steps > 0: t_total = args.max_steps args.num_train_epochs = args.max_steps // ( len(train_dataset) // args.gradient_accumulation_steps) + 1 else: t_total = len( train_dataset ) // 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 = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) 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 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) # multi-gpu training (should be after apex fp16 initialization) if args.n_gpu > 1: model = torch.nn.DataParallel(model) # Distributed training (should be after apex fp16 initialization) if args.local_rank != -1: model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True) args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu) # 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 * (torch.distributed.get_world_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) # 计算最大steps import math args.all_steps = math.ceil((len(train_dataset) * args.num_train_epochs / args.per_gpu_train_batch_size)) global_step = 0 epochs_trained = 0 steps_trained_in_current_epoch = 0 # # Check if continuing training from a checkpoint # if os.path.exists(args.model_name_or_path): # # set global_step to gobal_step of last saved checkpoint from model path # global_step = int(args.model_name_or_path.split("-")[-1].split("/")[0]) # 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) tr_loss, logging_loss = 0.0, 0.0 best_f1, n_no_improve = 0, 0 model.zero_grad() train_iterator = trange(int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0]) set_seed(args) # Added here for reproductibility for _ in train_iterator: stage_num = (len(train_dataset) // args.train_batch_size) + 1 epoch_iterator = trange(stage_num, desc="Iteration", disable=args.local_rank not in [-1, 0]) sub_train_dataset_list = [ train_dataset[i:i + args.train_batch_size] for i in range(0, len(train_dataset), args.train_batch_size) ] for step, __ in enumerate(epoch_iterator): sub_train_dataset = sub_train_dataset_list[step] sub_train_dataset = train_data_processor.get_tensor( sub_train_dataset) train_sampler = RandomSampler( sub_train_dataset ) if args.local_rank == -1 else DistributedSampler( sub_train_dataset) train_dataloader = DataLoader( sub_train_dataset, sampler=train_sampler, batch_size=args.train_batch_size, collate_fn=collate_fn, num_workers=args.num_workers, ) batch = train_dataloader.__iter__().__next__() # Skip past any already trained steps if resuming training if steps_trained_in_current_epoch > 0: steps_trained_in_current_epoch -= 1 continue model.train() batch = tuple(t.to(args.device) for t in batch) labels = batch[5] inputs = { "input_ids": batch[0], "input_modal": batch[2], "attention_mask": batch[1], "modal_start_tokens": batch[3], "modal_end_tokens": batch[4], } outputs = model(**inputs) logits = outputs[ 0] # model outputs are always tuple in transformers (see doc) loss = criterion(logits, labels) 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: logs = {} 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, criterion) for key, value in results.items(): eval_key = "eval_{}".format(key) logs[eval_key] = value loss_scalar = (tr_loss - logging_loss) / args.logging_steps learning_rate_scalar = scheduler.get_lr()[0] logs["learning_rate"] = learning_rate_scalar logs["loss"] = loss_scalar logging_loss = tr_loss for key, value in logs.items(): tb_writer.add_scalar(key, value, global_step) print(json.dumps({**logs, **{"step": global_step}})) if args.local_rank in [-1, 0] and args.save_steps > 0 and ( global_step % args.save_steps == 0 or global_step == args.all_steps): # Save model checkpoint output_dir = os.path.join( args.output_dir, "checkpoint-{}".format(global_step)) if not os.path.exists(output_dir): os.makedirs(output_dir) model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training torch.save(model_to_save.state_dict(), os.path.join(output_dir, WEIGHTS_NAME)) # 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) 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 # TODO 若训练超过规定的训练epochs的次数时,评估指标已不再提升,则停止训练。由于牵扯到evaluate,故先注释掉。 # if args.local_rank == -1: # result = evaluate(args, model, tokenizer, criterion) # if results["micro_f1"] > best_f1: # best_f1 = results["micro_f1"] # n_no_improve = 0 # else: # n_no_improve += 1 # # if n_no_improve > args.patience: # train_iterator.close() # break if args.local_rank in [-1, 0]: tb_writer.close() return global_step, tr_loss / global_step
def train(args, train_dataset, model, tokenizer): """ 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) train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset) train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size) 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"] # unfreeze_layers = ['layer.10', 'layer.11', 'bert.pooler', 'out.'] # unfreeze_layers = ['layer.10', 'layer.11', 'bert.pooler', 'out.'] # params = [[], []] # for name, param in model.named_parameters(): # param.requires_grad = False # for ele in unfreeze_layers: # if ele in name: # param.requires_grad = True # break # if not any(nd in name for nd in no_decay): # params[0].append(param) # else: # params[1].append(param) optimizer_grouped_parameters = [ { "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], # "params": params[0], "weight_decay": args.weight_decay, }, { "params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], # "params": params[1], "weight_decay": 0.0 }, ] optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) 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 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) # multi-gpu training (should be after apex fp16 initialization) if args.n_gpu > 1: model = torch.nn.DataParallel(model) # Distributed training (should be after apex fp16 initialization) if args.local_rank != -1: model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True, ) # 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 * (torch.distributed.get_world_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) # 计算最大steps import math args.all_steps = math.ceil((len(train_dataset) * args.num_train_epochs / args.per_gpu_train_batch_size)) global_step = 0 epochs_trained = 0 steps_trained_in_current_epoch = 0 # # Check if continuing training from a checkpoint # if os.path.exists(args.model_name_or_path): # # set global_step to gobal_step of last saved checkpoint from model path # global_step = int(args.model_name_or_path.split("-")[-1].split("/")[0]) # 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) tr_loss, logging_loss = 0.0, 0.0 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 reproductibility 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 model.train() batch = tuple(t.to(args.device) for t in batch) inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]} if args.model_type != "distilbert": inputs["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs = model(**inputs) 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: logs = {} 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(): eval_key = "eval_{}".format(key) logs[eval_key] = value loss_scalar = (tr_loss - logging_loss) / args.logging_steps learning_rate_scalar = scheduler.get_lr()[0] logs["learning_rate"] = learning_rate_scalar logs["loss"] = loss_scalar logging_loss = tr_loss for key, value in logs.items(): tb_writer.add_scalar(key, value, global_step) print(json.dumps({**logs, **{"step": global_step}})) if args.local_rank in [-1, 0] and args.save_steps > 0 and ( global_step % args.save_steps == 0 or global_step == args.all_steps): # Save model checkpoint output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step)) if not os.path.exists(output_dir): os.makedirs(output_dir) 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) 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
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 = RandomSampler( train_dataset) if args.local_rank == -1 else DistributedSampler( train_dataset) 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 = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) 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) # multi-gpu training (should be after apex fp16 initialization) if args.n_gpu > 1: model = torch.nn.DataParallel(model) # Distributed training (should be after apex fp16 initialization) if args.local_rank != -1: model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True) # 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 * (torch.distributed.get_world_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