Although Oracle does not have one of the best relationships with the open-source community, it recently made headlines with the launch of a new open-source tool dubbed Graphpipe.
The tool is made to standardize and simplify the deployment of machine learning (ML) models, what’s more, it boasts a collection of tools and libraries intended for following the standard.
Vish Abrams is the man charged with the task of spearheading the project. He has an impressive background, which includes assisting to create OpenStack at NASA and the launching of Nebula, an OpenStack startup, back in 2011.
Abrams said that they identified a gap as he and his team explored machine learning workflow further. Since teams spend a considerable amount of energy to create a machine learning model, it becomes difficult to deploy it for clients to use. Graphpipe can help remedy this situation.
READ MORE – 10 Applications of Deep Learning in Business
According to Abrams, it is typical for individuals to get absorbed into the hype when it comes to newer technologies such as machine learning. In fact, he emphasized his point by saying that people do not often consider the deployment process even though the development process is ever-improving.
Abrams told TechCrunch that Graphpipe represents the result of his team’s attempt to enhance the deployment efforts for machine learning models as well as establish an open standard governing the exercise in a bid to improve the space.
As Oracle delved deeper in the issue, they discovered three key impediments. First, there do not exist standard technique of serving APIs, which ends up leaving you to utilize whatever your framework offers.
Secondly, there’s no standard deployment approach, which causes developers to customize their own every time. Lastly, the company noted that for existing techniques, performance is an afterthought, which is not ideal for machine learning.
According to Abram, Oracle created Graphpipe to solve those three challenges. On the blog post revealing the details about the release of the open-source tool, he said it offers a standard, high-performance protocol intended for conveying tensor data through the network coupled with simple servers and clients’ implementations, which make querying and deploying of machine learning models from whatever framework seamless.
The blog post revealed that Oracle decides to create this standard and open-source it in an effort of trying to push the deployment of machine learning model forward.
According to Abram, Graphpipe rests on the intersection that exists between pushing the technology forward and solving business issues. For him, the ideal way to do that is to create an open-source technique.
Even though Abram recognized the tension that has been there for years between the open-source community and Oracle, he said that both parties have been collaborating to eliminate the perception. This effort can be shown by the contributions to the Oracle Functions Project and Kubernetes.
He went ahead to say that if the technology is interesting, individuals would give it a chance irrespective of the party behind its launch or creation.
Furthermore, Abram closed his statements by saying that Oracle cares more about the standard being widely adopted as compared to its implementation of it since that simplifies things for everybody.