# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Fine-tuning the library models for language modeling on a text file (GPT, GPT-2, BERT, RoBERTa). GPT and GPT-2 are fine-tuned using a causal language modeling (CLM) loss while BERT and RoBERTa are fine-tuned using a masked language modeling (MLM) loss. """ from __future__ import absolute_import import os import sys import pickle import torch import json import random import logging import argparse import numpy as np from io import open from itertools import cycle import torch.nn as nn from model import Seq2Seq from tqdm import tqdm, trange from bleu import _bleu from torch.utils.data import DataLoader, Dataset, SequentialSampler, RandomSampler,TensorDataset from torch.utils.data.distributed import DistributedSampler from transformers import (WEIGHTS_NAME, AdamW, get_linear_schedule_with_warmup, RobertaConfig, RobertaModel, RobertaTokenizer) MODEL_CLASSES = {'roberta': (RobertaConfig, RobertaModel, RobertaTokenizer)} logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s', datefmt = '%m/%d/%Y %H:%M:%S', level = logging.INFO) logger = logging.getLogger(__name__) class Example(object): """A single training/test example.""" def __init__(self, idx, source, target, ): self.idx = idx self.source = source self.target = target # def read_examples(filename): # """Read examples from filename.""" # examples=[] # with open(filename,encoding="utf-8") as f: # for idx,js in enumerate(json.load(f)): # source=' '.join(js['old_comment_tokens']) # target=' '.join(js['new_comment_tokens']) # examples.append( # Example( # idx = idx, # source=source, # target=target, # ) # ) # return examples def read_examples(filename): """Read examples from filename.""" examples=[] assert len(filename.split(','))==2 src_filename = filename.split(',')[0] trg_filename = filename.split(',')[1] idx = 0 with open(src_filename) as f1,open(trg_filename) as f2: for line1,line2 in zip(f1,f2): examples.append( Example( idx = idx, source=line1.strip(), target=line2.strip(), ) ) idx+=1 return examples class InputFeatures(object): """A single training/test features for a example.""" def __init__(self, example_id, source_ids, target_ids, source_mask, target_mask, ): self.example_id = example_id self.source_ids = source_ids self.target_ids = target_ids self.source_mask = source_mask self.target_mask = target_mask def convert_examples_to_features(examples, tokenizer, args,stage=None): features = [] for example_index, example in enumerate(examples): #source source_tokens = tokenizer.tokenize(example.source)[:args.max_source_length-2] source_tokens =[tokenizer.cls_token]+source_tokens+[tokenizer.sep_token] source_ids = tokenizer.convert_tokens_to_ids(source_tokens) source_mask = [1] * (len(source_tokens)) padding_length = args.max_source_length - len(source_ids) source_ids+=[tokenizer.pad_token_id]*padding_length source_mask+=[0]*padding_length #target if stage=="test": target_tokens = tokenizer.tokenize("None") else: target_tokens = tokenizer.tokenize(example.target)[:args.max_target_length-2] target_tokens = [tokenizer.cls_token]+target_tokens+[tokenizer.sep_token] target_ids = tokenizer.convert_tokens_to_ids(target_tokens) target_mask = [1] *len(target_ids) padding_length = args.max_target_length - len(target_ids) target_ids+=[tokenizer.pad_token_id]*padding_length target_mask+=[0]*padding_length if example_index < 5: if stage=='train': logger.info("*** Example ***") logger.info("idx: {}".format(example.idx)) logger.info("source_tokens: {}".format([x.replace('\u0120','_') for x in source_tokens])) logger.info("source_ids: {}".format(' '.join(map(str, source_ids)))) logger.info("source_mask: {}".format(' '.join(map(str, source_mask)))) logger.info("target_tokens: {}".format([x.replace('\u0120','_') for x in target_tokens])) logger.info("target_ids: {}".format(' '.join(map(str, target_ids)))) logger.info("target_mask: {}".format(' '.join(map(str, target_mask)))) features.append( InputFeatures( example_index, source_ids, target_ids, source_mask, target_mask, ) ) return features def _truncate_seq_pair(tokens_a, tokens_b,tokens_c, max_length): """Truncates a sequence pair in place to the maximum length.""" # This is a simple heuristic which will always truncate the longer sequence # one token at a time. This makes more sense than truncating an equal percent # of tokens from each, since if one sequence is very short then each token # that's truncated likely contains more information than a longer sequence. while True: total_length = len(tokens_a) + len(tokens_b)+len(tokens_c) if total_length <= max_length: break if len(tokens_a) >= len(tokens_b) and len(tokens_a)>=len(tokens_c): tokens_a.pop() elif len(tokens_b) >= len(tokens_a) and len(tokens_b)>=len(tokens_c): tokens_b.pop() else: tokens_c.pop() def set_seed(args): """set random seed.""" random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) if args.n_gpu > 0: torch.cuda.manual_seed_all(args.seed) def main(): parser = argparse.ArgumentParser() ## Required parameters parser.add_argument("--model_type", default=None, type=str, required=True, help="Model type: e.g. roberta") parser.add_argument("--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model: e.g. roberta-base" ) parser.add_argument("--tokenizer_name", default="", required=True, help="Pretrained tokenizer name or path if not the same as model_name") parser.add_argument("--output_dir", default=None, type=str, required=True, help="The output directory where the model predictions and checkpoints will be written.") parser.add_argument("--load_model_path", default=None, type=str, help="Path to trained model: Should contain the .bin files" ) ## Other parameters parser.add_argument("--train_filename", default=None, type=str, help="The train filenames (source and target files).") parser.add_argument("--dev_filename", default=None, type=str, help="The dev filename. (source and target files).") parser.add_argument("--test_filename", default=None, type=str, help="The test filename. (source and target files).") parser.add_argument("--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name") parser.add_argument("--max_source_length", default=64, type=int, help="The maximum total source sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.") parser.add_argument("--max_target_length", default=32, type=int, help="The maximum total target sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.") 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("--do_test", action='store_true', help="Whether to run eval on the dev set.") parser.add_argument("--do_lower_case", action='store_true', help="Set this flag if you are using an uncased model.") parser.add_argument("--no_cuda", action='store_true', help="Avoid using CUDA when available") parser.add_argument("--train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.") parser.add_argument("--eval_batch_size", default=8, 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("--beam_size", default=10, type=int, help="beam size for beam search") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight deay 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=3.0, 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("--eval_steps", default=-1, type=int, help="") parser.add_argument("--train_steps", default=-1, type=int, help="") parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument('--seed', type=int, default=42, help="random seed for initialization") # print arguments args = parser.parse_args() logger.info(args) # 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) torch.distributed.init_process_group(backend='nccl') args.n_gpu = 1 logger.warning("Process rank: %s, device: %s, n_gpu: %s, distributed training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1)) args.device = device # Set seed set_seed(args) # make dir if output_dir not exist if os.path.exists(args.output_dir) is False: os.makedirs(args.output_dir) config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path) tokenizer = tokenizer_class.from_pretrained(args.tokenizer_name,do_lower_case=args.do_lower_case) #budild model encoder = model_class.from_pretrained(args.model_name_or_path,config=config) decoder_layer = nn.TransformerDecoderLayer(d_model=config.hidden_size, nhead=config.num_attention_heads) decoder = nn.TransformerDecoder(decoder_layer, num_layers=6) model=Seq2Seq(encoder=encoder,decoder=decoder,config=config, beam_size=args.beam_size,max_length=args.max_target_length, sos_id=tokenizer.cls_token_id,eos_id=tokenizer.sep_token_id) if args.load_model_path is not None: logger.info("reload model from {}".format(args.load_model_path)) model.load_state_dict(torch.load(args.load_model_path)) model.to(device) if args.local_rank != -1: # Distributed training try: from apex.parallel import DistributedDataParallel as DDP except ImportError: raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use distributed and fp16 training.") model = DDP(model) elif args.n_gpu > 1: # multi-gpu training model = torch.nn.DataParallel(model) if args.do_train: # Prepare training data loader train_examples = read_examples(args.train_filename) train_features = convert_examples_to_features(train_examples, tokenizer,args,stage='train') all_source_ids = torch.tensor([f.source_ids for f in train_features], dtype=torch.long) all_source_mask = torch.tensor([f.source_mask for f in train_features], dtype=torch.long) all_target_ids = torch.tensor([f.target_ids for f in train_features], dtype=torch.long) all_target_mask = torch.tensor([f.target_mask for f in train_features], dtype=torch.long) train_data = TensorDataset(all_source_ids,all_source_mask,all_target_ids,all_target_mask) if args.local_rank == -1: train_sampler = RandomSampler(train_data) else: train_sampler = DistributedSampler(train_data) train_dataloader = DataLoader(train_data, sampler=train_sampler, batch_size=args.train_batch_size//args.gradient_accumulation_steps) num_train_optimization_steps = args.train_steps # 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=num_train_optimization_steps) #Start training logger.info("***** Running training *****") logger.info(" Num examples = %d", len(train_examples)) logger.info(" Batch size = %d", args.train_batch_size) logger.info(" Num epoch = %d", num_train_optimization_steps*args.train_batch_size//len(train_examples)) model.train() dev_dataset={} nb_tr_examples, nb_tr_steps,tr_loss,global_step,best_bleu,best_loss = 0, 0,0,0,0,1e6 bar = range(num_train_optimization_steps) train_dataloader=cycle(train_dataloader) eval_flag = True idx=0 for step in bar: batch = next(train_dataloader) batch = tuple(t.to(device) for t in batch) source_ids,source_mask,target_ids,target_mask = batch loss,_,_ = model(source_ids=source_ids,source_mask=source_mask,target_ids=target_ids,target_mask=target_mask) if args.n_gpu > 1: loss = loss.mean() # mean() to average on multi-gpu. if args.gradient_accumulation_steps > 1: loss = loss / args.gradient_accumulation_steps tr_loss += loss.item() train_loss=round(tr_loss*args.gradient_accumulation_steps/(nb_tr_steps+1),4) if (global_step + 1)%100==0: logger.info(" step {} loss {} batch-{}".format(global_step + 1,train_loss, ((global_step+1)*args.train_batch_size) / len(train_examples))) nb_tr_examples += source_ids.size(0) nb_tr_steps += 1 loss.backward() if (nb_tr_steps + 1) % args.gradient_accumulation_steps == 0: #Update parameters optimizer.step() optimizer.zero_grad() scheduler.step() global_step += 1 eval_flag = True if args.do_eval and ((global_step + 1) %args.eval_steps == 0) and eval_flag: #Eval model with dev dataset tr_loss = 0 nb_tr_examples, nb_tr_steps = 0, 0 eval_flag=False if 'dev_loss' in dev_dataset: eval_examples,eval_data=dev_dataset['dev_loss'] else: eval_examples = read_examples(args.dev_filename) eval_features = convert_examples_to_features(eval_examples, tokenizer, args,stage='dev') all_source_ids = torch.tensor([f.source_ids for f in eval_features], dtype=torch.long) all_source_mask = torch.tensor([f.source_mask for f in eval_features], dtype=torch.long) all_target_ids = torch.tensor([f.target_ids for f in eval_features], dtype=torch.long) all_target_mask = torch.tensor([f.target_mask for f in eval_features], dtype=torch.long) eval_data = TensorDataset(all_source_ids,all_source_mask,all_target_ids,all_target_mask) dev_dataset['dev_loss']=eval_examples,eval_data eval_sampler = SequentialSampler(eval_data) eval_dataloader = DataLoader(eval_data, sampler=eval_sampler, batch_size=args.eval_batch_size) logger.info("\n***** Running evaluation *****") logger.info(" Num examples = %d", len(eval_examples)) logger.info(" Batch size = %d", args.eval_batch_size) #Start Evaling model model.eval() eval_loss,tokens_num = 0,0 for batch in eval_dataloader: batch = tuple(t.to(device) for t in batch) source_ids,source_mask,target_ids,target_mask = batch with torch.no_grad(): _,loss,num = model(source_ids=source_ids,source_mask=source_mask, target_ids=target_ids,target_mask=target_mask) eval_loss += loss.sum().item() tokens_num += num.sum().item() #Pring loss of dev dataset model.train() eval_loss = eval_loss / tokens_num result = {'eval_ppl': round(np.exp(eval_loss),5), 'global_step': global_step+1, 'train_loss': round(train_loss,5)} for key in sorted(result.keys()): logger.info(" %s = %s", key, str(result[key])) logger.info(" "+"*"*20) #save last checkpoint last_output_dir = os.path.join(args.output_dir, 'checkpoint-last') if not os.path.exists(last_output_dir): os.makedirs(last_output_dir) model_to_save = model.module if hasattr(model, 'module') else model # Only save the model it-self output_model_file = os.path.join(last_output_dir, "pytorch_model.bin") torch.save(model_to_save.state_dict(), output_model_file) if eval_lossbest_bleu: logger.info(" Best bleu:%s",dev_bleu) logger.info(" "+"*"*20) best_bleu=dev_bleu # Save best checkpoint for best bleu output_dir = os.path.join(args.output_dir, 'checkpoint-best-bleu') if not os.path.exists(output_dir): os.makedirs(output_dir) model_to_save = model.module if hasattr(model, 'module') else model # Only save the model it-self output_model_file = os.path.join(output_dir, "pytorch_model.bin") torch.save(model_to_save.state_dict(), output_model_file) # 每一轮记录checkpoint if int((global_step+1)*args.train_batch_size / len(train_examples)) == idx+1: logger.info(" batch:%s",idx) output_dir = os.path.join(args.output_dir, 'epoch_{}'.format(idx+1)) if not os.path.exists(output_dir): os.makedirs(output_dir) model_to_save = model.module if hasattr(model, 'module') else model ckpt_output_path = os.path.join(output_dir, 'subject_model.pth') logger.info("Saving model checkpoint to %s", ckpt_output_path) torch.save(model_to_save.state_dict(), ckpt_output_path) idx = idx+1 if args.do_test: files=[] if args.dev_filename is not None: files.append(args.dev_filename) if args.test_filename is not None: files.append(args.test_filename) for idx,file in enumerate(files): logger.info("Test file: {}".format(file)) eval_examples = read_examples(file) eval_features = convert_examples_to_features(eval_examples, tokenizer, args,stage='test') all_source_ids = torch.tensor([f.source_ids for f in eval_features], dtype=torch.long) all_source_mask = torch.tensor([f.source_mask for f in eval_features], dtype=torch.long) eval_data = TensorDataset(all_source_ids,all_source_mask) # Calculate bleu eval_sampler = SequentialSampler(eval_data) eval_dataloader = DataLoader(eval_data, sampler=eval_sampler, batch_size=args.eval_batch_size) model.eval() p=[] for batch in tqdm(eval_dataloader,total=len(eval_dataloader)): batch = tuple(t.to(device) for t in batch) source_ids,source_mask= batch with torch.no_grad(): preds = model(source_ids=source_ids,source_mask=source_mask) for pred in preds: t=pred[0].cpu().numpy() t=list(t) if 0 in t: t=t[:t.index(0)] text = tokenizer.decode(t,clean_up_tokenization_spaces=False) p.append(text) model.train() predictions=[] accs=[] with open(os.path.join(args.output_dir,"test_{}.output".format(str(idx))),'w') as f, open(os.path.join(args.output_dir,"test_{}.gold".format(str(idx))),'w') as f1: for ref,gold in zip(p,eval_examples): predictions.append(str(gold.idx)+'\t'+ref) f.write(ref+'\n') f1.write(gold.target+'\n') accs.append(ref==gold.target) dev_bleu=round(_bleu(os.path.join(args.output_dir, "test_{}.gold".format(str(idx))).format(file), os.path.join(args.output_dir, "test_{}.output".format(str(idx))).format(file)),2) logger.info(" %s = %s "%("bleu-4",str(dev_bleu))) logger.info(" %s = %s "%("xMatch",str(round(np.mean(accs)*100,4)))) logger.info(" "+"*"*20) if __name__ == "__main__": main()