参考：
https://zhuanlan.zhihu.com/p/60286685
1、创建一个文件夹 ,比如a_docker
2、进入a_docker创建三个文件
3、requirements.txt 文件
bert4keras==0.9.9
Flask==1.1.1
4、Dockfile 文件
FROM ubuntu:16.04
FROM python:3.6.5
RUN apt-get update -y && \
apt-get install -y python-pip python-dev
# We copy just the requirements.txt first to leverage Docker cache
COPY ./requirements.txt /app/requirements.txt
WORKDIR /app
RUN apt-get install vim
RUN pip install -r requirements.txt
COPY . /app
CMD python /app/server.py
5、server.py 文件
from flask import Flask
app = Flask(__name__)
@app.route('/')
def hello_world():
return 'Hello World!'
if __name__ == '__main__':
app.run("0.0.0.0", 6600, threaded=True)
6、在a_docker当前文件夹目录下打开cmd进行打包
docker build -t simbert:v1 .
7、打包结束后运行镜像
docker run -p 8089:6600 -d 打包后的镜像id
8、进入容器id安装tensorflow
docker exec -it 容器id /bin/bash
pip install tensorflow==1.14 -i https://pypi.douban.com/simple
9、退出容器，更改后的容器再打包更新生成镜像
docker commit -m="rm bertcheckpoint" 容器id simbert:v3
10、阿里云镜像服务创建个仓库
https://cr.console.aliyun.com/cn-hangzhou/instance/repositories
11、push到阿里云
$ docker login --username=108373****@qq.com registry.cn-hangzhou.aliyuncs.com
$ docker tag [ImageId] registry.cn-hangzhou.aliyuncs.com/loongxl/simbert:[镜像版本号]
$ docker push registry.cn-hangzhou.aliyuncs.com/loongxl/simbert:[镜像版本号]
docker login --username=108***[email protected] registry.cn-hangzhou.aliyuncs.com
docker tag 2faa9***c registry.cn-hangzhou.aliyuncs.com/loongxl/simbert:v5
docker push registry.cn-hangzhou.aliyuncs.com/loongxl/simbert:v5
12、下载拉取镜像
docker pull registry.cn-hangzhou.aliyuncs.com/loongxl/simbert:v5
13、运行镜像
docker run -p 8089:6600 -d 镜像id
curl localhost:8089
文件挂载
参考：https://zhuanlan.zhihu.com/p/87593614
linux版本：
:/loong不能是/app 因为会覆盖app里内容，/loong会在容器内显示对应宿主机上的loong文件
sudo docker run -p 8089:6600 -v /tmp/loong:/loong -d 2faa98ff038c
windows版本：
/D表示d盘
docker run -p 8089:6600 -v /D/t***rt_docker/chinese_simbert_L-6_H-384_A-12:/loong -d 2faa98ff038c
修改server.py文件，加载宿主机文件11.txt（/loong/11.txt是容器里的目录，映射的是宿主机的11.txt）
实时更改11.txt文件内容，web访问刷新也会改变
更改server.py在线更新镜像部署模型
进入原容器内更改
server.py
import numpy as np
import os
from collections import Counter
os.environ['TF_KERAS'] = '1'
from bert4keras.backend import keras, K
from bert4keras.models import build_transformer_model
from bert4keras.tokenizers import Tokenizer
from bert4keras.snippets import sequence_padding
from bert4keras.snippets import uniout
from keras.models import Model
from flask import Flask
from flask import render_template, request, jsonify
import tensorflow as tf
maxlen = 32
sess=keras.backend.get_session()
graph=tf.get_default_graph()
# bert配置
config_path = 'loong/bert_config.json'
checkpoint_path = 'loong/bert_model.ckpt'
dict_path = 'loong/vocab.txt'
# 建立分词器
tokenizer = Tokenizer(dict_path, do_lower_case=True) # 建立分词器
# 建立加载模型
bert = build_transformer_model(
config_path,
checkpoint_path,
with_pool='linear',
application='unilm',
return_keras_model=False,
)
encoder = keras.models.Model(bert.model.inputs, bert.model.outputs[0])
app = Flask(__name__)
@app.route('/', methods=['GET', 'POST'])
def build_plot():
query = request.args.get('query')
token_ids = tokenizer.encode(query, maxlen=maxlen)[0]
#在默认会话与计算图中进行模型的预测
with graph.as_default():
with sess.as_default():
a_vecs = encoder.predict([a_token_ids, np.zeros_like(a_token_ids)],
verbose=True)
a_vecs = a_vecs / (a_vecs**2).sum(axis=1, keepdims=True)**0.5
return jsonify({"hah":a_vecs})
if __name__ == '__main__':
app.run("0.0.0.0", 6600, debug=True, threaded=True)
更新新镜像并运行
docker commit -m="rm bertcheckpoint" 019728c1e196 simbert:v10
docker run -p 8099:6600 -v /D/t***-12_test:/loong -d 950f56fe6556