TensorFlow Serving是通过rpc接口远程访问tensorflow引擎的服务器。
- TensorFlow Serving文档 https://tensorflow.github.io/serving/architecture_overview
TensorFlow Serving可以通过容器启动,方法如下:
docker run --name tfserver --restart always tensorflow/tf_grpc_server:0.8.0 \
--cluster_spec="worker|localhost:2220;localhost:2221,ps|localhost:3220;localhost:3221" \
--job_name=worker --task_id=0
-
详细的TensorFlow Serving容器构建方法参考: https://github.com/openthings/serving/blob/master/tensorflow_serving/g3doc/setup.md
-
在分布式环境中运行TensorFlow的详细指南: https://www.tensorflow.org/how_tos/distributed/#distributed-tensorflow