RHEL vs. RHEL AI: What's the difference?

At Red Hat Summit 2024, Red Hat Enterprise Linux (RHEL) AI was unveiled. RHEL AI is described as, "a foundation model platform that enables users to more seamlessly develop, test and run generative AI (gen AI) models," but that description might not be super clear to everyone, especially RHEL users accustomed to the platform’s role as a comprehensive hybrid cloud operating system.

In this article, I describe the two offerings in detail, including how RHEL and RHEL AI are different, how they're the same, what RHEL AI includes, and how you can get started with RHEL AI today.

What is RHEL?

Red Hat Enterprise Linux is the world’s leading enterprise Linux platform and it provides a consistent operating foundation for modern IT and enterprise hybrid cloud deployments. And RHEL is more than just an operating system. Every RHEL subscription includes:

• Enterprise-grade automation and management tools, including Red Hat Insights
• Built-in security features like live kernel patching, security profiles, security standards certification and a software supply chain rooted in trusted code
• Access to Red Hat's extensive partner ecosystem
• 24x7 technical support
• A 10 year product life cycle
• Legal protections provided by our Open Source Assurance program
• …and more

What is RHEL AI?

In the simplest terms, RHEL AI is RHEL. More specifically, RHEL AI is an optimized and highly-tuned version of RHEL that provides a foundation model platform for consistently developing, refining, testing and running Granite family large language models (LLM) to power enterprise AI applications.

RHEL AI also takes advantage of image mode for Red Hat Enterprise Linux, a new approach to operating system deployment that lets users build, deploy and manage RHEL as a bootable container (bootc) image. This means RHEL AI can be deployed, managed and scaled the same way you deploy, manage and scale any containerized application.

In essence, RHEL AI is a fully-equipped LLM development, tuning and hosting platform that can be deployed and managed with Red Hat OpenShift AI anywhere across the hybrid cloud, from on-prem to edge.

What does RHEL AI include?

RHEL AI combines a number of components, including:

• Granite open source-licensed LLMs that are distributed under the Apache 2.0 license with industry-leading transparency into training datasets and model weights
• A supported, lifecycled distribution of InstructLab that provides a scalable, cost-effective solution for enhancing LLM capabilities and making knowledge and skills contributions much more accessible to people who do not have the expertise normally needed for AI fine-tuning
• A bootable image of RHEL (enabled by image mode for RHEL) which includes popular AI libraries such as PyTorch, and hardware-optimized accelerators for NVIDIA, Intel and AMD
• Enterprise-grade technical support provided by Red Hat
• Open Source Assurance legal protections available to all Red Hat customers with active, paid Red Hat software subscriptions

What are the Granite large language models?

The Granite family of open source LLMs were developed by IBM and include Code, Time Series, Language and GeoSpatial models. These have been made available on Hugging Face under a permissive Apache 2.0 license that enables broad, unencumbered commercial use.

RHEL AI currently includes the collaboratively-developed Granite models from the InstructLab community. In the initial release of RHEL AI, users have GA access to the granite-7b-starter model and the corresponding granite-7b-redhat-lab model, and technology previews of the granite-8b-code-base and granite-8b-code-instruct Code models.  You can get an up-to-date list and learn more about the LLMs included with RHEL AI in the official documentation.

Importantly, all open source Granite models are trained on carefully curated data, with industry-leading levels of transparency into the data and model weights used.

Why is data transparency important?

While gen AI has seen a massive spike in use and popularity, many companies are hesitant to use third-party LLMs due to concerns about the output they produce and whether it can be legally used.

For example, the New York Times has sued OpenAI with the claim that chatGPT was trained on over a million copyrighted NYTimes articles. As of September 2024 that has not been proven in a court of law, but if it is, it could mean that any output generated by chatGPT could be considered a derivative work, meaning the NYTimes would own the copyright on that work.

Obviously, situations like these are a clear concern for anyone using gen AI tools.

Granite open source LLMs are different, however, as they are released under an Apache 2.0 license and include significant transparency into the training data and model weights. This means that users can directly confirm what data has been used to train the models, reducing the risk of legal conflicts or related issues in the future. Additionally, Red Hat provides legal protections to Red Hat customers using RHEL AI through the Open Source Assurance program.

What is InstructLab?

Started by IBM and Red Hat, the InstructLab project's goal is to democratize gen AI through the power of an open source community. InstructLab is focused on building simple and cost-effective tools for fine-tuning LLMs used in gen AI applications.

These tools are designed to be more accessible to developers and subject matter experts who have vital expertise but who may not have the specialized technical knowledge usually required for fine-tuning AI models. This means more people are able to participate in these fine-tuning processes, making them more democratic, efficient and effective.

The core of the InstructLab project is LAB: Large-Scale Alignment for ChatBots, described in detail in this 2024 research paper by members of the MIT-IBM Watson AI Lab and IBM Research.

Getting started with InstructLab for generative AI model tuning

What is the Open Source Assurance program?

Through the Open Source Assurance program, Red Hat offers intellectual property assurances in the event of a claim that open source software or an AI model provided by Red Hat violates a third party's intellectual property rights.

This means that, in the event a third party brings a claim for infringement, Red Hat will defend the customer against the claim and pay the amount of a judgment or settlement subject to stated limitations. Red Hat may also replace or modify the software or AI model in question, where possible.

All customers with active, paid Red Hat software subscriptions are eligible for the Open Source Assurance program, which is an integral benefit of a Red Hat product subscription.

Red Hat has offered legal protections before

The Open Source Assurance program isn't new. It was originally launched in 2004 to go along with Red Hat's patent promise and our participation as a founding member of the Open Invention Network.

Later, in 2006, the Open Source Assurance program was expanded to include these more sweeping intellectual property-related legal protections for Red Hat customers. These protections were originally offered in relation to the Linux kernel, which was the center of some controversial litigation at the time.

The program is intended to assure customers that claims against Red Hat products  are an issue for Red Hat to resolve, not the customer. This means customers can use Red Hat solutions without interruption in the event of an intellectual property infringement claim.

Including RHEL AI and the output of Red Hat-branded AI models in the Open Source Assurance program is simply building on Red Hat's long history of protecting customers who use our open source solutions.

Ready to get started with RHEL AI?

Red Hat Developers has introduced a RHEL AI learning path that will help you explore these new tools and capabilities. This no-cost training introduces the fundamentals related to data collection, pre-training and fine-tuning and walks you through the process of enhancing an existing LLM.

Try Red Hat Enterprise Linux AI