General
What is Iguazio and what are the core benefits of the platform?
Iguazio is an AI and gen AI platform that is used to accelerate time to market for developing business services and applications that are based on machine learning models. With Iguazio, you can develop, deploy, monitor, and manage AI and gen AI applications at scale. Data science, data engineering and DevOps teams can automate the entire ML pipeline from data management and model development to rapid deployment and on-going LiveOps.
What are the main components of your architecture?
There are 2 main components, data nodes and application nodes. Data nodes require flash/nvme drives and host the control plane for our environment. The application nodes run Kubernetes and host all the services you deploy.
Who are the primary users for the platform?
The primary users are AI engineers, data scientists, and data engineers.
Do you integrate with other ML platforms?
Yes, we can integrate with other ML platforms such as Sagemaker and Azure ML.
Do you support GPUs?
Yes for both training and serving models. GPU sharing is a big advantage of running on our platform.
Where can your solution be deployed?
Iguazio can be deployed on AWS, Azure, Google cloud, or on-prem. In the cloud, it is usually deployed on the customers’ VPC. The platform can also run in an air-gapped environment (no internet connectivity). For more information see Deployment and Specifications.
Can I auto-scale to increase the size of my cluster for the duration of a job?
Yes, we support auto-scaling through managed K8s services such as EKS, AKS, and GKE. This also includes the ability to define node groups with specific sizes, machine types, spot vs. on-demand instance type, etc. See Best Practices.
Am I limited to working only within your platform?
No. Services and data you define in Iguazio can be accessed outside the platform. Following the same level of security as within the platform, developers can write code using their favorite tools.
Do you provide pre-built models and/or AutoML?
Iguazio is an AI/ML development platform. We include all the tools a developer needs to write the code to train models and run inferencing. We currently provide pre-built libraries, out of the box, that can assist you.
What is the relationship between the Iguazio Platform, MLRun, and Nuclio?
The Iguazio Platform is an enterprise AI platform. MLRun and Nuclio are open source software which are used in Iguazio's enterprise platform. MLRun is an orchestration tool for quickly building and managing continuous (gen) AI applications across their lifecycle. Nuclio is a serverless function framework. It is used by MLRun as an engine for deploying online serving functions, running ML models, or for real time data ingestion. For more information see the documentation for MLRun and Nuclio.
What is an MLRun function?
An MLRun function is an abstraction that allows you to run a piece of code on top of a Kubernetes cluster without worrying about Kubernetes. It also allows you to specify configuration details such as cluster resources, environment variables, Docker images + pip installs, batch/distributed/real-time runtimes, and more, using high level Python syntax.
What's the difference between an MLRun function and a real-time Nuclio function and when should each be used?
An MLRun function is an abstraction that allows you to easily run a piece of code on top of a Kubernetes cluster with configuration, runtimes, resources, etc. One of the configuration options is to specify the runtime, for example, a regular K8s job, Spark/Dask/Horovod distributed runtime, or real-time Nuclio function.
A real-time Nuclio function is simply one of the available runtimes. You would choose Nuclio for anything involving real-time, such as model serving or real-time data ingestion.
For more information see MLRun Overview.
Can we create function templates that can only be used within the organization?
Yes, you can create re-usable function templates by leveraging MLRun functions. These can be stored in a private code repository such as GitHub. In addition, each job is versioned, and you can specify the version of the function you want to run.
What resources and skills are needed to manage your environment?
We can offer Managed Services deployment where our team handles the operations of the frameworks. If you decide to deploy on your own environment, the required knowledge matches that of a Systems Administrator with DevOps and Kubernetes experience. It is a very low maintenance environment.
Does Iguazio offer a Jupyter service?
Yes, every data scientist can work on their own Jupyter service running on Iguazio, however it's recommended to use your own IDE (see Setting up your Environment).