Announcing the General Availability of Red Hat Enterprise Linux AI (RHEL AI) Version 1.2!

We are excited to announce the general availability of RHEL AI 1.2, the foundation model platform designed to help organizations develop, fine-tune, deploy, and run open-source Granite Large Language Models (LLMs) to power enterprise applications. With RHEL AI, developers can now seamlessly fine-tune Red Hat and IBM-produced Granite foundational models to meet their specific needs. RHEL AI 1.2 builds on the success of previously available features, such as the indemnified Granite LLMs and the supported InstructLab workflow for model alignment, enhancing the overall experience for developers with support for wider hardware and public clouds. The images for RHEL AI 1.2 are available here.

What’s new in RHEL AI 1.2?

1. Expanded Hardware Support:
   1. RHEL AI 1.2 now supports AMD accelerators. Visit Hardware Support for further information.
   2. Technology Preview is now available for Google Cloud Platform (GCP) and Azure with 8xA100 and 8xH100 accelerators for the full end-to-end InstructLab Model Alignment workflow and Model Inferencing.
   3. Continued support for NVIDIA accelerated computing continues to be supported on bare metal and AWS and IBM cloud
   4. Support for Lenovo ThinkSystem SR675 V3 servers including factory preload option
   5. Support for AMD Instinct Accelerators (technology preview)
2. Enhanced Cloud Availability: With BYOS (bring your own subscriptions), in addition to existing support for AWS and IBM Cloud, you can now install RHEL AI 1.2 on GCP and Azure as a technology preview. This makes RHEL AI even more accessible for a wider range of cloud infrastructures.
3. InstructLab Tools and Workflow Enhancements:
   1. RHEL AI’s InstructLab workflow—used for fine-tuning models by adding new knowledge—is now generally available on IBM Cloud, Google Cloud and Azure alongside bare metal and AWS.
   2. Hardware auto-detection now simplifies the setup process, and the new --training-journal flag enables users to continue previously failed training runs, making model fine-tuning more efficient and resilient.
   3. Training checkpoint and resume: Long training runs during model fine tuning can now be saved at regular intervals, thanks to periodic checkpointing. This feature allows InstructLab users to resume training from the last saved checkpoint instead of starting over, saving valuable time and computational resources.
   4. Enhanced training with PyTorch FSDP (technology preview): For multi-phase training of models with synthetic data, ilab train now uses PyTorch Fully Sharded Data Parallel (FSDP). This dramatically reduces training times by sharding a model’s parameters, gradients and optimizer states across data parallel workers (e.g., GPUs). Users can pick FSDP for their distributed training by using ilab config edit.
4. Synthetic Data Generation (SDG): The enhanced LAB synthetic data generation (SDG) allows you to create large artificial datasets with advanced multi-phase training strategies, enabling more accurate and efficient model training.
5. Model Inferencing Support: Model inferencing with vLLM, a memory-efficient engine, is now available as a technology preview on GCP along with previously supported generally availability for bare metal, AWS, and IBM Cloud.
6. Supported LLMs: granite-7b-starter, granite-7b-redhat-lab, mixtral-8x7B-instruct-v0-1, and prometheus-8x7b-v2.0 models continue to be generally available (GA). The granite-8b-code-instruct and granite-8b-code-base models, designed for code generation, remain in technology preview.

RHEL AI 1.2 represents a significant step forward in fine-tuning enterprise-grade LLMs and expands deployment options across leading public cloud providers with expanded infrastructure support for both NVIDIA and AMD. The enhanced hardware support, robust synthetic data generation, and improved tooling ensure your organization can tailor AI models to meet specific enterprise needs efficiently.

For more details, check out the full release notes and explore all the new features and enhancements! For further information and a detailed understanding of what you can do with RHEL AI, visit RHEL AI Overview. 

Important notice:

With the introduction of RHEL AI 1.2, we will be deprecating support for RHEL AI 1.1 in 30 days. Please ensure your systems are upgraded to RHEL AI 1.2 to continue receiving support.