update to transformers

README.md

@@ -2,7 +2,7 @@
 
 ## Description
 
-- Chinese version of GPT2 training code, using BERT tokenizer or BPE tokenizer. It is based on the extremely awesome repository from HuggingFace team [Pytorch-Transformers](https://github.com/huggingface/pytorch-transformers). Can write poems, news, novels, or train general language models. Support char level, word level and BPE level. Support large training corpus.
+- Chinese version of GPT2 training code, using BERT tokenizer or BPE tokenizer. It is based on the extremely awesome repository from HuggingFace team [Transformers](https://github.com/huggingface/transformers). Can write poems, news, novels, or train general language models. Support char level, word level and BPE level. Support large training corpus.
 - 中文的GPT2训练代码，使用BERT的Tokenizer或GPT2自带的BPE Tokenizer或Sentencepiece的BPE model（感谢[kangzhonghua](https://github.com/kangzhonghua)的贡献）。可以写诗，新闻，小说，或是训练通用语言模型。支持字为单位或是分词模式或是BPE模式。支持大语料训练。
 - 微信交流群：请见Issue第一条。
 

eval.py

@@ -1,4 +1,4 @@
-import pytorch_transformers
+import transformers
 import torch
 import os
 import json
@@ -65,7 +65,7 @@ def main():
 
     os.environ["CUDA_VISIBLE_DEVICES"] = args.device  # 此处设置程序使用哪些显卡
 
-    model_config = pytorch_transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
+    model_config = transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
     print('config:\n' + model_config.to_json_string())
 
     n_ctx = model_config.n_ctx
@@ -97,7 +97,7 @@ def main():
         print('you need to specify a trained model.')
         exit(1)
     else:
-        model = pytorch_transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
+        model = transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
     model.eval()
     model.to(device)
 

generate.py

@@ -3,7 +3,7 @@ import torch.nn.functional as F
 import os
 import argparse
 from tqdm import trange
-from pytorch_transformers import GPT2LMHeadModel
+from transformers import GPT2LMHeadModel
 
 
 def is_word(word):

generate_texts.py

@@ -1,10 +1,9 @@
 import torch
 import torch.nn.functional as F
-import pytorch_transformers
 import os
 import argparse
 from tqdm import trange
-from pytorch_transformers import GPT2LMHeadModel
+from transformers import GPT2LMHeadModel
 
 os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3"  # 此处设置程序使用哪些显卡
 
@@ -139,7 +138,6 @@ def main():
     device = "cuda" if torch.cuda.is_available() else "cpu"
 
     tokenizer = tokenization_bert.BertTokenizer(vocab_file=args.tokenizer_path)
-    model_config = pytorch_transformers.GPT2Config.from_json_file(args.model_config)
     model = GPT2LMHeadModel.from_pretrained(args.model_path)
     model.to(device)
     model.eval()

requirements.txt

@@ -1,4 +1,4 @@
-pytorch-transformers
+transformers
 torch
 numpy
 tqdm

tokenizations/tokenization_bert.py

@@ -22,7 +22,7 @@ import os
 import unicodedata
 from io import open
 
-from pytorch_transformers.tokenization_utils import PreTrainedTokenizer
+from transformers.tokenization_utils import PreTrainedTokenizer
 
 logger = logging.getLogger(__name__)
 

train.py

@@ -1,4 +1,4 @@
-import pytorch_transformers
+import transformers
 import torch
 import os
 import json
@@ -78,7 +78,7 @@ def main():
 
     os.environ["CUDA_VISIBLE_DEVICES"] = args.device  # 此处设置程序使用哪些显卡
 
-    model_config = pytorch_transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
+    model_config = transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
     print('config:\n' + model_config.to_json_string())
 
     n_ctx = model_config.n_ctx
@@ -118,9 +118,9 @@ def main():
         print('files built')
 
     if not args.pretrained_model:
-        model = pytorch_transformers.modeling_gpt2.GPT2LMHeadModel(config=model_config)
+        model = transformers.modeling_gpt2.GPT2LMHeadModel(config=model_config)
     else:
-        model = pytorch_transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
+        model = transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
     model.train()
     model.to(device)
 
@@ -139,8 +139,8 @@ def main():
     total_steps = int(full_len / stride * epochs / batch_size / gradient_accumulation)
     print('total steps = {}'.format(total_steps))
 
-    optimizer = pytorch_transformers.AdamW(model.parameters(), lr=lr, correct_bias=True)
-    scheduler = pytorch_transformers.WarmupLinearSchedule(optimizer, warmup_steps=warmup_steps,
+    optimizer = transformers.AdamW(model.parameters(), lr=lr, correct_bias=True)
+    scheduler = transformers.WarmupLinearSchedule(optimizer, warmup_steps=warmup_steps,
                                                           t_total=total_steps)
     if fp16:
         try:

train_single.py

@@ -1,4 +1,4 @@
-import pytorch_transformers
+import transformers
 import torch
 import os
 import json
@@ -69,7 +69,7 @@ def main():
         from tokenizations import tokenization_bert
 
     os.environ["CUDA_VISIBLE_DEVICES"] = args.device  # 此处设置程序使用哪些显卡
-    model_config = pytorch_transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
+    model_config = transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
     print('config:\n' + model_config.to_json_string())
 
     n_ctx = model_config.n_ctx
@@ -101,9 +101,9 @@ def main():
         print('files built')
 
     if not args.pretrained_model:
-        model = pytorch_transformers.modeling_gpt2.GPT2LMHeadModel(config=model_config)
+        model = transformers.modeling_gpt2.GPT2LMHeadModel(config=model_config)
     else:
-        model = pytorch_transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
+        model = transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
     model.train()
     model.to(device)
     multi_gpu = False
@@ -115,8 +115,8 @@ def main():
     total_steps = int(full_len / stride * epochs / batch_size / gradient_accumulation)
     print('total steps = {}'.format(total_steps))
 
-    optimizer = pytorch_transformers.AdamW(model.parameters(), lr=lr, correct_bias=True)
-    scheduler = pytorch_transformers.WarmupLinearSchedule(optimizer, warmup_steps=warmup_steps,
+    optimizer = transformers.AdamW(model.parameters(), lr=lr, correct_bias=True)
+    scheduler = transformers.WarmupLinearSchedule(optimizer, warmup_steps=warmup_steps,
                                                           t_total=total_steps)
     if fp16:
         try: