当前位置:网站首页>Oracle announces open source graphpipe, a new standard of machine learning model

Oracle announces open source graphpipe, a new standard of machine learning model

2021-07-20 04:10:04 mob604756fb8908


# Scan the QR code above to sign up for Chengdu Yuanchuang Hui #

License agreement :UPL

development language :C/C++

operating system : Cross platform

Developer :Oracle

This high performance standard protocol should make it easier for enterprises to deploy and query machine learning models from any framework .

Machine learning is expected to change the status quo of the industry . however , Its application speed in enterprises is slower than expected , Because it's hard for these organizations to deploy and manage machine learning technologies themselves . Part of the challenge is that machine learning models often use customization techniques for training and deployment , So it is difficult to deploy models across servers or different departments .

So , Oracle hopes to transmit tensor data through open source and high-performance standard network protocol (tensor data) —— Such a technical means to solve the above challenges . This new standard , Oracle bone inscriptions call it GraphPipe, It makes it easier for enterprises to deploy and query machine learning models from any framework .

The official response to GraphPipe The explanation of is , It's a collection of protocols and software , It aims to simplify the deployment of machine learning model and separate it from the implementation of framework specific model .

GraphPipe Designed to address three special challenges :


First , Model services API There is no standard , This means that business applications often need custom clients to communicate with deployed models .


Next , It's very difficult to build a model server , And there are few out of the box deployment solutions .


Last , The solution that enterprises usually use now ( Such as python-JSON API) Unable to deliver the performance required for business critical applications .

GraphPipe Include


A group of flatbuffer Definition


according to flatbuffer Guidelines for defining consistent models


Examples of models from various machine learning frameworks


Used by GraphPipe Query the client library of the model

GraphPipe features


be based on flatbuffers Micro machine learning transmission specification


Apply to Tensorflow,Caffe2 and ONNX Simple and efficient reference model server (reference model servers)


Go,Python and Java Efficient client implementation based on XML

Use these tools , Enterprises should be able to deploy models across multiple servers , Or create model collections from different frameworks using common protocols .GraphPipe It can help to deploy machine learning for IOT applications that rely on remote running models .

版权声明
本文为[mob604756fb8908]所创,转载请带上原文链接,感谢
https://chowdera.com/2021/05/20210530142032027G.html

随机推荐