nvidia-docker安装：https://github.com/NVIDIA/nvidia-docker

nvidia-docker的基本操作：https://blog.csdn.net/pandamax/article/details/79151107

TensorFlow Serving + Docker整体流程：https://zhuanlan.zhihu.com/p/52096200

更多例子：

https://zhuanlan.zhihu.com/p/64413178

https://zhuanlan.zhihu.com/p/58474810

https://github.com/pakrchen/text-antispam

keras模型转tfserving：

https://www.jianshu.com/p/91aae37f1da6

model端Dockfile的编写

FROM  tensorflow/serving:1.13.0-gpu 
MAINTAINER XXX
COPY model /models/model

#docker build -t  image_name .
#NV_GPU=0 nvidia-docker run -p 8500:8500 -t image_name --model_config_file=/path/to/models.config  modelfs.config主要保存模型的名称，路径，model_platform

tfserving warmup 

其中tf2.0代码已经改为tf.io.TFRecordWriter

https://blog.csdn.net/hzhj2007/article/details/97679342