1. 简介
BERT 的核心过程非常简洁，它会先从数据集抽取两个句子，其中第二句是第一句的下一句的概率是 50%，这样就能学习句子之间的关系 。其次随机去除两个句子中的一些词，并要求模型预测这些词是什么，这样就能学习句子内部的关系 。最后再将经过处理的句子传入大型模型，并通过两个损失函数同时学习上面两个目标就能完成训练 。
2 运行(只有CPU) 2.1 运行环境
1.官方推荐.10以上的版本，可以搭建tf1.12
2.从gitee上下载整个项目，同时需要将仓库中Data目录下的-12_H-768_A-12文件夹和tmp文件夹一并加入到这个项目中去 。
接下来就要开始预训练了
【bert使用指南】2.2 修改自己的文件路径名
在args.py下
""和’’修改为自己的tmp文件夹的路径即可
2.3 运行
python create_pretraining_data.py
得到类似如下输出
INFO:tensorflow:*** Example ***INFO:tensorflow:tokens: [CLS] indeed , it was recorded in [MASK] star that a fortunate early [MASK] ##r had once picked up on the highway a solid chunk [MASK] gold quartz which the [MASK] had freed from its inc [MASK] ##ing soil , and washed into immediate and [MASK] popularity . [SEP] rainy season , [MASK] insult show habit of body , and seldom lifted their eyes to the rift ##ed [MASK] india - ink washed skies [MASK] them . " cass " beard [MASK] elliot early that morning , but not with a view to [MASK] . a leak in his [MASK] roof , - - quite [MASK] with his careless , imp ##rov ##ide ##nt habits , - - had rouse ##d him at 4 a [MASK] m [SEP]INFO:tensorflow:input_ids: 101 5262 1010 2009 2001 2680 1999 103 2732 2008 1037 19590 2220 103 2099 2018 2320 3856 2039 2006 1996 3307 1037 5024 20000 103 2751 20971 2029 1996 103 2018 10650 2013 2049 4297 103 2075 5800 1010 1998 8871 2046 6234 1998 103 6217 1012 102 16373 2161 1010 103 15301 2265 10427 1997 2303 1010 1998 15839 4196 2037 2159 2000 1996 16931 2098 103 2634 1011 10710 8871 15717 103 2068 1012 1000 16220 1000 10154 103 11759 2220 2008 2851 1010 2021 2025 2007 1037 3193 2000 103 1012 1037 17271 1999 2010 103 4412 1010 1011 1011 3243 103 2007 2010 23358 1010 17727 12298 5178 3372 14243 1010 1011 1011 2018 27384 2094 2032 2012 1018 1037 103 1049 102INFO:tensorflow:input_mask: 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1INFO:tensorflow:segment_ids: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1INFO:tensorflow:masked_lm_positions: 7 12 13 25 30 36 45 52 53 54 68 74 81 82 93 99 103 105 125 0INFO:tensorflow:masked_lm_ids: 17162 2220 4125 1997 4542 29440 20332 4233 1037 16465 2030 2682 2018 13763 5456 6644 1011 8335 1012 0INFO:tensorflow:masked_lm_weights: 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 0.0INFO:tensorflow:next_sentence_labels: 0INFO:tensorflow:*** Example ***INFO:tensorflow:tokens: [CLS] and there burst on phil ##am ##mon ' s astonished eyes a vast semi ##ci ##rcle of blue sea [MASK] ring ##ed with palaces and towers [MASK] [SEP] like most of [MASK] fellow gold - seekers , cass was super ##sti [MASK] . [SEP]INFO:tensorflow:input_ids: 101 1998 2045 6532 2006 6316 3286 8202 1005 1055 22741 2159 1037 6565 4100 6895 21769 1997 2630 2712 103 3614 2098 2007 22763 1998 7626 103 102 2066 2087 1997 103 3507 2751 1011 24071 1010 16220 2001 3565 16643 103 1012 102 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0INFO:tensorflow:input_mask: 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0INFO:tensorflow:segment_ids: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0INFO:tensorflow:masked_lm_positions: 10 20 23 27 32 39 42 0 0 0 0 0 0 0 0 0 0 0 0 0INFO:tensorflow:masked_lm_ids: 22741 1010 2007 1012 2010 2001 20771 0 0 0 0 0 0 0 0 0 0 0 0 0INFO:tensorflow:masked_lm_weights: 1.0 1.0 1.0 1.0 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0INFO:tensorflow:next_sentence_labels: 1INFO:tensorflow:Wrote 60 total instances