If you’re headed to Red Hat Summit, May 19-22, join engineers from Red Hat Research and Emerging Technologies for labs and breakout sessions that let you get your hands on what they’ve been doing and dive deep into far edge and AI/LLM topics. Click on a presentation title to go to the Summit website and add it to your schedule (schedule is subject to change).
If you’re not going to make it to Red Hat Summit in Boston this year, check out the links following the abstracts and stay connected to what’s next by following the Emerging Technologies and Red Hat Research blogs, and by subscribing to the Red Hat Research Quarterly.
Language models and AI
Evaluating and building Triton GPU kernels in an AI/LLM stack
Learn how to tune and benchmark your LLM inference deployment to maximize performance and optimize costs by focusing on GPU kernels and Triton.Triton is a Python-based DSL compiler and related tooling designed for writing efficient GPU kernels in a hardware-agnostic manner, offering high-level abstractions while enabling low-level performance optimization for AI and HPC workloads. Optimizing the inference performance of an LLM can make a big difference in the cost-effectiveness of an enterprise LLM deployment.
Join Steven Royer, Principal Software Engineer, Sanjeev Rampal, Senior Principal Software Engineer, and Craig Magina, Principal Software Engineer, for this breakout session. We’ll explore practical optimization techniques such as:
- Setting up a development environment for writing Triton kernel
- Balancing memory and compute trade-offs.
- Profiling popular LLM models to identify optimizations for your use case
- Optimizing GPU kernel efficiency
Additionally, we’ll discuss portability options to help you get the most from your AI hardware investments.
Learn more:
- Democratizing AI accelerators and GPU kernel programming using Triton: Sanjeev Rampal provides an overview of the Triton language, foundational architecture topics in the Triton space, and the Triton ecosystem.
- Getting started with PyTorch and Triton on AMD GPUs using the Red Hat Universal Base Image: Sajeev Rampal and Steven Royer provide guidelines for getting started with developing and using Triton kernels, in combination with or independent of frameworks such as PyTorch, and doing so using Red Hat Universal Base Image/Fedora-based containers, and while leveraging the AMD Instinct family of AI accelerators.
- A container-first approach to Triton development: Maryam Tahhan describes the benefits of containerized Triton development, including simplified setup, greater consistency, and more seamless collaboration.
Vibe code a data-driven website with generative AI for smart aquaculture in Red Hat OpenShift AI
In this hands-on lab, we’ll develop a new smart aquaculture web platform with data models. You’ll learn fine-grained access control with the Red Hat Build of Keycloak. You’ll build FIWARE Smart Data Models for Smart Aquaculture with generative AI in a Red Hat OpenShift AI Visual Studio Code (VSCode) Java workbench. With this platform, we’ll visualize fish and population data to help mitigate food insecurity.
Vibe coding with the Computate open source code generation platform begins with using machine learning to parse and index your code into a search engine (including Java classes, constructors, methods, and fields, as well as dependent parent projects). A simple service watches your source code directory for new files and edits to code. Machine learning kicks in again and parses your latest code changes. Artificial intelligence can then query your code in the search engine, read your code comments, and generate additional code you are missing based on natural language descriptions of desired functionality. By writing 93 lines of code and descriptive code comments, vibe coding can save you from writing 10,000+ lines of code across multiple repositories.
The Smart Aquaculture project—a collaboration between Green Reef Foundation, Oceans 97, and Qatar Strategic Business Group—supports food processing in developing nations through workforce training with academic and industry partners using innovative digital twins, unmanned aerial vehicles (UAVs), and autonomous aerial vehicles (AAVs). Arise Fish and Seafood Process and the Green Reef Foundation combine high-fidelity simulations with digital culture preservation and training on open source edge computing to support these digital twins. Join Dewane Branch, CEO and founder of Qatar Strategic Business Group, Chris Tate, Red Hat Principal Software Engineer, and Jarvis Green, Founder of the Green Reef Foundation, to learn more about:
- Intelligent application programming interface (API) and dashboard code generation with the Computate platform
- Smart device registration with FIWARE microservices
- Event-driven subscriptions to smart device data changes
- Granular data access control between edge and cloud with Red Hat build of Keycloak
- Custom Leaflet dashboards for visualizing live GeoJSON data on maps
Learn more:
- Computate and smart aquaculture use case: Chris Tate provides an overview of working with FIWARE to develop open source-driven solutions to global problems and demonstrates how to deploy and experiment with your own IoT smart platform.
- Building better communities with streaming data and machine learning using FIWARE: Chris explains how FIWARE enabled work on Smart Villages technology, including a smart traffic light.
Signing AI models with Sigstore
Much like signing software, signing AI models protects model integrity and establishes that the model came from an authentic, trusted source. This helps prevent attacks like model or dataset poisoning and unauthorized modifications. Part of the Open Source Security Foundation, Sigstore’s Model Transparency project is a community project aimed at applying the software supply chain security practice of signing to machine learning models.
Join Red Hat Principal Software Engineer Lily Sturmann for this lightning talk to find out how this project is applying and updating secure signing practices to the AI ecosystem and how this could benefit your organization.
Learn more:
- Model authenticity and transparency with Sigstore: Ivan Font discusses why traditional signing does not work well for ML models and how the Sigstore model transparency project works.
Cloud-edge continuum
Manage fleets of edge devices and applications with Flight Control
Flight Control is a project for declaratively and securely managing fleets of edge devices and their applications. Users declare a fleet’s target configuration and rollout policy, and Flight Control takes care of rolling out changes to all devices in the fleet and rolling up status.Flight Control is designed for modern, container-centric toolchains and operational best practices. It works best on image-based Linux operating systems running bootc or ostree and with container workloads running on Podman/Docker or Kubernetes.
Join Avishay Traeger, Senior Principal Software Engineer, and Frank Zdarsky, Distinguished Engineer, for this breakout session and discover more about this project.
Learn more:
- Project vision: Flight Control documentation introduces key concepts and provides a high-level view of Flight Control architecture.
Project Jumpstarter: An open source framework for cloud-native hardware-in-the-loop testing in automotive DevSecOps
The automotive industry’s transition to software-defined vehicles (SDVs) has created new IT challenges, particularly developing software that targets specific electronic control units (ECUs) where hardware-in-the-loop (HIL) testing is required. In this lab, we’ll introduce Jumpstarter, an open source framework that supports cloud-native hardware-in-the-loop testing for automotive ECUs and other edge devices. Integrated with modern continuous integration and continuous delivery (CI/CD) tools like Red Hat OpenShift Pipelines, Jumpstarter supports the latest DevSecOps practices—including intelligent automation and Infrastructure as Code (IaC)—while addressing the unique complexities of SDV development.
Join Miguel Angel Ajo Pelayo, Red Hat Senior Principal Software Engineer, Kirk Brauer, Hyundai America Technical Center, Inc. (HATCI) Infotainment Software Engineer, and Michael Kuehl, Red Hat Principal Specialist Solution Architect, for real-world examples and live demonstrations that show how Jumpstarter integrates with Red Hat Developer Hub to accelerate development cycles, improve quality assurance, and support easier collaboration across the automotive software development lifecycle.
Discover how Jumpstarter is transforming SDV development by:
- Supporting cloud-native HIL testing that scales across distributed development teams
- Automating complex ECU test scenarios within existing CI/CD workflows
- Providing IaC capabilities for hardware test environments
Learn more:
- Jumpstarter: Enabling hardware in the loop: Miguel Angel Ajo Pelayo presents Jumpstarter at DevConf.CZ 2024 and explains how it makes embedded software testing easier and more efficient.
- Jumpstarter: Cloud-native hardware in the loop: Nick Cao, Red Hat Software Engineer, discusses the architecture and implementation of Jumpstarter at DevConf.US 2024 and explains how it fits into the hardware-in-the-loop ecosystem.
Horizon Europe projects working on the cloud edge continuum
In this lightning talk, Jose Castillo Lema, Performance Engineer, and Luis Tomas Bolivar, Principal Software Engineer, will explore some of the work by Red Hat researchers and engineers within projects funded by Horizon Europe, the EU’s key program for research and innovation. We’ll discuss three projects currently being undertaken and present some of the interesting innovations being developed, including a developer-focused multicluster scheduler, an edge device controller, a multitechnology network operator, and more.
Learn more:
- AI, edge-cloud, and service/network management research underway at Red Hat Ireland: Dr. Leigh Griffin, Red Hat Senior Research Manager Ireland, introduces the Horizon Europe projects: AC3, CODECO, and INCODE (now P2CODE).
- Edge and cloud computing conference spotlights CODECO decentralized edge-cloud orchestration: Red Hat Principal Software Engineer Ricardo Noriega de Soto presented an overview of the edge computing landscape and a demo highlighting the potential of AI featuring CODECO.
