Building A Platform for Machine Learning Operations from Open Source Frameworks
Yan Liu, Zhijing Ling, Boyu Huo, Boqian Wang, Tianen Chen, Esma Mouine
Abstract
Machine Learning Operations (MLOps) aim to establish a set of practices that put tools, pipelines, and processes to build fast time-to-value machine learning development projects. The lifecycle of machine learning project development encompasses a set of roles, stacks of software frameworks and multiple types of computing resources. Such complexity makes MLOps support usually bundled with commercial cloud platforms that is referred as vendor lock. In this paper, we provide an alternative solution that devises a MLOps platform with open source frameworks on any virtual resources. Our MLOps approach is driven by the development roles of machine learning models. The tool chain of our MLOps connects to the typical CI/CD workflow of machine learning applications. We demonstrate a working example of training and deploying a machine learning model for the application of detecting software repository code vulnerability.