You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
TensorFlow Serving includes two policies that accommodate most known use- cases.
These are the Availability Preserving Policy (avoid leaving zero versions loaded;
typically load a new version before unloading an old one), and the Resource Preserving Policy
(avoid having two versions loaded simultaneously, thus requiring double the resources;
unload an old version before loading a new one)
支持从多种存储上加载模型:
可以扩展支持更多种类的存储。
client端访问的批处理功能:
同样,这个功能也是可以自定义policy。
Batching of multiple requests into a single request can significantly reduce the cost
of performing inference, especially in the presence of hardware accelerators such as GPUs.
TensorFlow serving
首先是调研了一下
TensorFlow Serving
。他的arch view
文档在这里 或者这里。除了提供基础的
rpc server
的功能外,亮点在于一下几个feature
:load
多个版本的model
,并且客户端可以访问指定的版本。model
发布后,自动加载新版本。Availability Preserving Policy
和Resource Preserving Policy
。client
端访问的批处理功能:根据文档描述,其
Loaders
是可以扩展的,这样具有了支持非TensorFlow model
的能力。社区已经有人为TensorFlow Serving
增加caffe
模型的支持:tensorflow/serving#261https://github.com/rayglover-ibm/serving-caffe
厂内情况
TensorFlow Serving
做了一些,不过仅支持TensorFlow
的模型。baidurpc
做一个简单Infer Server
。讨论:
我们在大会之前做一个可以和
TensorFlow Serving
对标的Serving
服务作为亮点,时间有点来不及。TensorFlow Serving
的基础上做插件支持PaddlePaddle
的模型?Infer Server
(如HttpServer)?The text was updated successfully, but these errors were encountered: