The Kubeflow community is rapidly growing due to its contributions to advancing AI by streamlining the AI/ML experience in Kubernetes. Kubeflow provides a composable ecosystem for implementing end-to-end solutions for AI/ML. Kubeflow includes the following projects: Trainer, Central Dashboard, Pipelines, Notebooks, Katib, KServe, Model registry, and Spark Operator.
At KubeCon + CloudNativeCon, end-users, contributors, and distributors meet to share knowledge, use cases, best practices, and their passion for technology and how this brings solutions to businesses and different types of organizations.
Kubeflow presence during Keynotes
Several organizations mentioned Kubeflow to showcase AI/ML solutions
- From Oracle to bring F1 Simulation with Oracle Red Bull Racing using Kubernetes and Kubeflow as part of the stack to run their trackside, factory and simulations on a highly scalable, reliable and optimized platform.
- Vodafone’s Perspective to show The Future of CloudNative in Telco
- The journey of cloud-native adoption in Telecom provides ongoing challenges and available solutions in creating an open telco cloud infrastructure
- A real-time sign language interpretation by Rob Koch.
- It was an engaging and insightful session that showcased a real use case where Kubeflow provides AI/ML tools to remove communication barriers and empower accessibility through Kubernetes.
Kubeflow Summit
Kubeflow summit was full of energy from product announcements, use cases, and community unique experiences were shared on a half-day event.
The Kubeflow Summit schedule offered diverse sessions for all levels. Starting with Kubeflow’s high-level overview of the product, roadmap, and future, the use of JAX for Distributed tracing and hyperparameter Optimization to Arrow Data caching, Distributed edge, Profiles Automation, and more.
Cloud Native + Kubernetes AI Day
Panel: Engaging the Kubeflow Community: Building an Enterprise-Ready AI/ML Platform
Speakers: Yuan Tang, Red Hat; Andrey Velichkevich, Kubeflow Steering Committee; Andreea Munteanu, Canonical; Johnu George, Nutanix; Ronen Dar, NVIDIA
The panelist shared the challenges of being an official distribution of such a product,customer use cases, and the influence they had over the project’s roadmap.Additionally, this panel highlights the importance of open governance and company diversity in the Kubeflow community.
Kubeflow Release 1.10
Kubeflow Release Manager Ricardo Martinelli de Oliveira announced the Release of Kubeflow Platform 1.10. Thanks to the community and everyone who contributed to this release.
Main features of Kubeflow Release 1.10:
- New UI for Model Registry
- Spark Operator as a core Kubeflow component
- Kubernetes and container security (CISO compatibility)
- Hyperparameter Optimization for LLMs Fine-Tuning
- Loop parallelism in Pipelines
- New parameter distributions for Katib
Explore the full release details
KubeCon + CloudNativeCon sessions
During KubeCon + CloudNativeCon, Kubeflow had many sessions across different tracks, observability, Generative AI, and the community maintainer session. These sessions were engaging and diverse, including panels, breakout sessions, and tutorials. Huge thanks to the speakers who contributed to the success of KubeCon + CloudNativeCon and the Kubeflow use cases.
Maintainer track: Kubeflow Ecosystem: What’s Next for Cloud Native AI/ML and LLMOps
The speakers highlighted Kubeflow’s components and the community around them. Kubeflow as a solution for GenAI applications to build scalable and secure solutions that run distributed across any Kubernetes cluster. Showcasing pipeline orchestration and data processing to distributed training, tuning, and inference.
Speakers: Andrey Velichkevich Apple, Johnu George Nutanix,Yuan Tang Red Hat, Yuki Iwai CyberAgent, Valentina Rodriguez Sosa Red Hat
From High Performance Computing To AI Workloads on Kubernetes: MPI Runtime in Kubeflow TrainJob
This talk introduced the Kubeflow MPI Runtime integrated with Kubeflow Trainer 2.0 and Kubeflow TrainJob, enabling distributed training with MLX and LLMs fine-tuning using DeepSpeed on Kubernetes. It marks the first public presentation at KubeCon + CloudNativeCon to showcase distributed MLX running on Kubernetes.
Speakers: Andrey Velichkevich Apple, Yuki Iwai CyberAgent Inc
Project Pavilion
The Kubeflow community is grateful for having this opportunity to connect with many contributors, end-users, and new attendees. Attendees had the chance to interact with the Kubeflow community in person through technical and use case discussions, including technical deep dives with demos.
Learn about Proposal: Kubeflow Data Cache for Distributed Training on Kubernetes
Watch the demo:Speed up Your ML Workloads With Kubernetes Powered In-memory Data… Rasik Pandey & Akshay Chitneni
What’s next for Kubeflow
- Kubeflow Roadmap: High-level features
- Trainer 2.0 (including new CRDs, Kubeflow Python SDK, Custom dataset and model initialize, LLMs Fine-Tuning blueprints)
- Pipelines 2.5 (Security and UI improvements)
- Model Registry (For e.g. Model Integration Storage, Model Catalog)
- Spark Operator
- Kubeflow Data Cache with Apache Arrow
- CNCF Graduation
The Kubeflow community is preparing for CNCF Graduation. If you want to get involved, please join our Graduation calls.
Explore the Project details
- A new Kubeflow Outreach Committee
- Kubeflow New Working Groups and Proposals
Explore the many proposals and working groups that will be taking place this year. Click here
How to get involved with Kubeflow