1. BERT

BERT工具能够快速的得到词向量表示。名称叫做: bert-as-service，只要调用该服务就能够得到我们想要的向量表示。

2. 安装 bert-as-service

1） 环境要求:

Python版本 >= 3.5，Tensorflow版本 >= 1.10
本地环境配置：Win10 ，Python = 3.7 ，Tensorflow = 1.13.1
服务器环境配置：centos ，Python = 3.6 ，Tensorflow = 1.13.1

2）安装服务端和客户端

执行

pip install -U bert-serving-server bert-serving-client

（pip install -i https://pypi.tuna.tsinghua.edu.cn/simple -U bert-serving-server bert-serving-client）

3. 启动 BERT 服务

1）下载预训练模型

github :https://github.com/google-research/bert/

• BERT-Large, Uncased (Whole Word Masking): 24-layer, 1024-hidden, 16-heads, 340M parameters
• BERT-Large, Cased (Whole Word Masking): 24-layer, 1024-hidden, 16-heads, 340M parameters
• BERT-Base, Uncased: 12-layer, 768-hidden, 12-heads, 110M parameters
• BERT-Large, Uncased: 24-layer, 1024-hidden, 16-heads, 340M parameters
• BERT-Base, Cased: 12-layer, 768-hidden, 12-heads , 110M parameters
• BERT-Large, Cased: 24-layer, 1024-hidden, 16-heads, 340M parameters
• BERT-Base, Multilingual Cased (New, recommended): 104 languages, 12-layer, 768-hidden, 12-heads, 110M parameters
• BERT-Base, Multilingual Uncased (Orig, not recommended) (Not recommended, use Multilingual Casedinstead): 102 languages, 12-layer, 768-hidden, 12-heads, 110M parameters
• BERT-Base, Chinese: Chinese Simplified and Traditional, 12-layer, 768-hidden, 12-heads, 110M parameters

下载BERT-Base, Chinese模型解压，放在根目录下

2）启动服务

解压缩后，运行如下命令进行启动，目录换成解压后的路径。(-num_worker指定使用多少个CPU)

  bert-serving-start -model_dir /Users/mantch/Downloads/chinese_L-12_H-768_A-12 -num_worker=4

运行后会看到如下结果:

           ckpt_name = bert_model.ckpt
         config_name = bert_config.json
                cors = *
                 cpu = False
          device_map = []
       do_lower_case = True
  fixed_embed_length = False
                fp16 = False
 gpu_memory_fraction = 0.5
       graph_tmp_dir = None
    http_max_connect = 10
           http_port = None
        mask_cls_sep = False
      max_batch_size = 256
         max_seq_len = 25
           model_dir = ./chinese_L-12_H-768_A-12
no_position_embeddings = False
    no_special_token = False
          num_worker = 4
       pooling_layer = [-2]
    pooling_strategy = REDUCE_MEAN
                port = 5555
            port_out = 5556
       prefetch_size = 10
 priority_batch_size = 16
show_tokens_to_client = False
     tuned_model_dir = None
             verbose = False
                 xla = False

I:WORKER-1:[__i:gen:559]:ready and listening!
I:WORKER-0:[__i:gen:559]:ready and listening!
I:WORKER-2:[__i:gen:559]:ready and listening!
I:WORKER-3:[__i:gen:559]:ready and listening!
I:VENTILATOR:[__i:_ru:164]:all set, ready to serve request!

port = 5555，port_out = 5556 为端口号信息，启动成功。

3)调用 bert 进行自己的应用

from bert_serving.client import BertClient
bc = BertClient(ip='localhost',check_version=False, check_length=False)
vec = bc.encode(['学习'])
print(vec)

提示：

liunx刚开始装的Python版本与本地相同，出现各种报错
错误1：ImportError: /lib64/libstdc++.so.6: version `CXXABI_1.3.8' not found (required by /root/anaconda3/en

错误2：ImportError: /lib64/libm.so.6: version `GLIBC_2.23' not found (required by /root/anaconda3/envs/tensorflow11/lib/python3.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so)

因此将Python版本改为3.6，重新安装Tensorflow，启动服务成功。