💡Concepts
- Explore the philosophy behind Fiat
To understand why we start the Fiat project, we should know why we use MLOps.
Challenges of Machine Learning Workflow
Developing machine learning (ML) systems can be challenging and complex. Unlike traditional software development, ML development brings along unique challenges. We can frequently hear the same concerns -
There are various tools. Hundreds of open-source tools cover each phase of the ML lifecycle, from data preparation to model training. However, unlike traditional software development, where teams select one tool for each phase, in ML, you usually want to try every available tool (e.g., algorithm) to see whether it improves results. ML developers thus need to use and produce dozens of libraries.
It's hard to track experiments. Machine learning algorithms have dozens of configurable parameters. Whether you work alone or on a team, it isn't easy to follow which parameters, code, and data went into each experiment to produce a model.
It's hard to reproduce results. Teams often have trouble getting the same code to work again without detailed tracking. Whether you are a data scientist passing your training code to an engineer for use in production or going back to your past work to debug a problem, reproducing steps of the ML workflow is critical.
It's hard to deploy ML. Moving a model to production can be challenging due to the plethora of deployment tools and environments it needs to run in (e.g., REST serving, batch inference, or mobile apps). There is no standard way to move models from any library to any of these tools, creating a new risk with each new deployment.
Conventional AI application development methods require a lot of human resource costs, and the production lines of these models are highly coupled, non-reusable, and challenging to expand and maintain.
It is intuitive to grab the idea of constructing an artificial intelligence development platform that integrates AI algorithms, computing power, and power builder, providing corresponding workflows for functional modules such as machine learning, deep learning, and training models to overcome these challenges.
At the same time, the platform also provides computing power support required for development, enabling developers to effectively use the artificial intelligence capabilities in the platform to develop artificial intelligence products through interface calls.
Concept of MLOps
With Machine Learning Model Operationalization Management (MLOps), we want to provide an end-to-end machine learning development process to design, build and manage reproducible, testable, and evolvable ML-powered software.
Being an emerging field, MLOps is rapidly gaining momentum amongst Data Scientists, ML Engineers, and AI enthusiasts. The Continuous Delivery Foundation SIG MLOps differentiates ML model management from traditional software engineering and suggests the above MLOps capabilities.
Status Quo of MLOps Industry
Leading technology companies have developed internal MLOps platforms to accelerate the AI application development life cycle. These platforms are then packaged into PaaS products and sold to customers. AWS SageMaker, GCP VertexAI, Huawei ModelArts, and Baidu PaddlePaddle BML are some of the most popular.
Unfortunately, these platforms are expensive, making them inaccessible to students, labs, and individual developers. An open-source toolkit would be beneficial to assist in the model development life cycle using existing computing resources. Small companies and individual developers cannot afford these commercial platforms and the resources bound to them, but they still require a reliable framework to build self-hosted MLOps platforms.
🥳 That's how "Fiat" is proposed!
Given the circumstances mentioned above, we propose Fiat. This innovative and comprehensive open-source MLOps platform allows for developing AI applications through a microservice architecture, meeting the needs of developers and organizations looking to bring AI applications to market.
Fiat uses the latest stack of techniques and toolkits to provide the necessary tools and resources to simplify the AI development process. As a result, the platform makes creating sophisticated AI applications to meet modern businesses' demands easier. Furthermore, with its microservice architecture, Fiat is highly scalable, flexible, and cost-effective.
Moreover, Fiat is an open-source platform encouraging collaboration and sharing among developers and AI enthusiasts. The platform can be continuously updated and improved with new features and functions. It ensures that users always have access to the latest tools, frameworks, and resources and can stay at the forefront of AI development.
Reference
Last updated