将训练好的bert向量提供sentence 编码服务

1、参考

   https://zhuanlan.zhihu.com/p/50582974
   https://github.com/hanxiao/bert-asservice/blob/master/client/README.md

2、下载bert中文词向量

  地址:https://github.com/google-research/bert#pre-trained-models
  中文向量链接:https://storage.googleapis.com/bert_models/2018_11_03/chinese_L-12_H-768_A-12.zip

3、提供服务

3.1 将上述压缩文件解压
3.2 构建环境
安装依赖环境:
pip install numpy
pip install -U bert-serving-server[http]
pip install bert-serving-client
pip install tensorflow>=1.10.0

4、 提供本地/远程服务

4.1 本地直接调用:
     from bert_serving.client import BertClient
     bc = BertClient()
     bc.encode(['我 喜欢 你们', '我 喜 欢 你 们'])
 4.2  远程请求服务
     post服务:
      curl -X POST http://**.*.*.68:8125/encode -H 'content-type: application/json' -d '{"id": 123,"texts": ["hello world"], "is_tokenized": false}'

       返回结果:
            {
                "id":123,
                 "result":[[-0.00980051327496767,0.05821939557790756,-0.06836936622858047,
                                      -0.4723478853702545,0.48761454224586487,-1.4105712175369263, 
                                      ...
                                      ...
                                      ,-0.10073700547218323,-0.17246723175048828]],
                "status":200
            }
            
            
4.3、在一个GPU服务器(**.*.*.68)上部署bert服务,在另外一台cpu服务器(**.*.*.67)调用这个服务:
        step1: 调用前先在(**.*.*.68)上安装client:
              pip install bert-serving-client
        
        step2: 调用服务demo
            # on another CPU machine
            from bert_serving.client import BertClient
            bc = BertClient(ip='xx.xx.xx.xx')  # ip address of the GPU machine
            bc.encode(['First do it', 'then do it right', 'then do it better'])

5、 模型不需要分词,距离如下,发现这集中情况得到的编码向量是一样的

4289471-b1207a6ec80341c3.png
image.png

6、QA : https://github.com/hanxiao/bert-as-service/blob/master/client/README.md#speech_balloon-faq

猜你喜欢

转载自blog.csdn.net/weixin_33826268/article/details/86904180