Questo contenuto non è disponibile nella lingua selezionata.
Chapter 1. Red Hat Enterprise Linux AI 1.2 release notes
RHEL AI provides organizations with a process to develop enterprise applications on open source Large Language Models (LLMs).
1.1. About this release
Red Hat Enterprise Linux AI version 1.2 includes various features for Large Language Model (LLM) fine-tuning on the Red Hat and IBM produced Granite model. A customized model using the RHEL AI workflow consisted of the following:
- Install and launch a RHEL 9.4 instance with the InstructLab tooling.
- Host information in a Git repository and interact with a Git-based taxonomy of the knowledge you want a model to learn.
- Run the end-to-end workflow of synthetic data generation (SDG), multi-phase training, and benchmark evaluation.
- Serve and chat with the newly fine-tuned LLM.
1.2. Features and Enhancements
Red Hat Enterprise Linux AI version 1.2 includes various features for Large Language Model (LLM) fine-tuning.
1.2.1. Installing
Red Hat Enterprise Linux AI is installable as a bootable image. This image contains various tooling for interacting with RHEL AI. The image includes: Red Hat Enterprise Linux 9.4, Python version 3.11 and InstructLab tools for model fine-tuning. For more information about installing Red Hat Enterprise Linux AI, see Installation overview.
Red Hat Enterprise Linux AI version 1.2 continues installation options for bare metal, AWS and IBM Cloud. You can see all the supported installation options on RHEL AI in "Installation feature tracker". For more information about the hardware requirements for these platforms, see Red Hat Enterprise Linux AI hardware requirements.
1.2.1.1. Installing RHEL AI on systems with AMD accelerators (Technology Preview)
On RHEL AI version 1.2, you can now install and deploy Red Hat Enterprise Linux AI on machines with AMD accelerators as a technology preview. RHEL AI currently only supports AMD hardware on bare metal and Azure. For more information about RHEL AI hardware requirements for AMD, see Red Hat Enterprise Linux AI hardware requirements.
RHEL AI verion 1.2 currently does not provide a training profile for AMD hardware. You need to manually configure your config.yaml
file and add the proper training configurations for training on AMD. For the documentation on how to manually configure your AMD training profile, see Configuring the training profile for AMD accelerators (Technology preview).
1.2.1.2. Installing RHEL AI on Google Cloud Platform (GCP) (Technology Preview)
On RHEL AI version 1.2, you can now install and deploy Red Hat Enterprise Linux AI on Google Cloud Platform (GCP) instances as a technology preview. For the documentation on installing Red Hat Enterprise Linux AI on GCP, see Installing on Google Cloud Platform (GCP)
RHEL AI currently supports 8xA100 and 8xH100 accelerators on GCP instances for the full end-to-end workflow. You can also serve LLMs provided by Red Hat for inferencing on GCP instances. For more details on the RHEL AI hardware requirements for GCP, see Red Hat Enterprise Linux AI hardware requirements. For more information about the support scope of Red Hat Technology Preview features, see Technology Preview Features Support Scope.
1.2.1.3. Installing RHEL AI on Azure (Technology Preview)
You can now install and deploy Red Hat Enterprise Linux AI version 1.2 on Microsoft Azure as a technology preview. For the documentation on installing Red Hat Enterprise Linux AI on Azure, see Installing on Azure
RHEL AI currently supports 8xA100 and 8xH100 accelerators on Azure instances for the full end-to-end workflow. You can also serve LLMs provided by Red Hat for inferencing on Azure instances. For more details on the RHEL AI hardware requirements for AWS, see Red Hat Enterprise Linux AI hardware requirements. For more information about the support scope of Red Hat Technology Preview features, see Technology Preview Features Support Scope.
1.2.2. Building your RHEL AI environment
After installing Red Hat Enterprise Linux AI, you can set up your RHEL AI environment with the InstructLab tools.
1.2.2.1. Initializing InstructLab
You can initialize and set up your RHEL AI environment by running the ilab config init
command. This command creates the necessary configurations for interacting with RHEL AI and fine-tuning models. It also creates proper directories for your data files.
1.2.2.1.1. Hardware auto-detection
Red Hat Enterprise Linux AI version 1.2 now offers hardware auto-detection when initializing InstructLab. The CLI prompts you to confirm if the auto-detection selected your hardware correctly, then it automatically adds the training parameters to your config.yaml
file. For more information about hardware auto-detection, see the Initialize InstructLab documentation.
1.2.2.2. Downloading Large Language Models
You can download various Large Language Models (LLMs) provided by Red Hat to your RHEL AI machine or instance. You can download these models from a Red Hat registry after creating and logging in to your Red Hat registry account. For more information about the supported RHEL AI LLMs, see the Downloading models documentation and the "Large Language Models (LLMs) technology preview status".
The granite-8b-code-instruct
and granite-8b-code-base
code models are continuing in Technology Preview on RHEL AI version 1.2. For more information about the support scope of Red Hat Technology Preview features, see Technology Preview Features Support Scope.
1.2.2.3. Serving and chatting with models
Red Hat Enterprise Linux AI version 1.2 allows you to run a vLLM inference server on various LLMs. The vLLM tool is a memory-efficient inference and serving engine library for LLMs that is included in the RHEL AI image. For more information about serving and chatting with models, see Serving and chatting with the models documentation.
There are various networking endpoint customizations you can create on Red Hat Enterprise Linux AI. Including creating a API enpoint for serving a model, setting up your machine as a inference server, and various other options. For more information about these customizations, see Serving and chatting with the models documentation.
1.2.3. Customizing a Large Language Model (LLM) on RHEL AI
Red Hat Enterprise Linux AI allows you to customize and fine-tune the granite-7b-starter
base model with the RHEL AI end-to-end workflow.
1.2.3.1. Running the end-to-end workflow on IBM Cloud
On Red Hat Enterprise Linux AI version 1.2, you can now use IBM cloud to customize the Granite model and run the end-to-end InstructLab workflow. For more details on the RHEL AI end-to-end hardware requirements for IBM Cloud, see Red Hat Enterprise Linux AI hardware requirements.
1.2.3.2. Adding knowledge data to a Granite LLM.
On Red Hat Enterprise Linux AI, you can customize your taxonomy tree so a model can learn domain-specific information. You host your knowledge data in a Git repository and fine-tune a model with that data. In the RHEL AI workflow, you create a qna.yaml
file that includes questions and answers for the model to learn. This file gets run through the synthetic data generation (SDG) process, training, and evaluation, to then create a new LLM that contains the data from the Git repository and qna.yaml
file. For detailed documentation on how to create a knowledge markdown and YAML file, see Adding knowledge to your taxonomy tree.
1.2.3.3. Synthetic Data Generation (SDG)
Red Hat Enterprise Linux AI includes the LAB enhanced method of synthetic data generation (SDG). You can use the qna.yaml
files with your own knowledge data to create hundreds of artifical datasets in the SDG process. For more information about running the SDG process, see Generating a new dataset with Synthetic data generation (SDG).
1.2.3.4. Training a model with your data
Red Hat Enterprise Linux AI includes the LAB enhanced method of multi-phase training: A fine-tuning strategy where datasets are trained and evaluated in multiple phases to create the best possible model. For more details on multi-phase training, see Training your data on the model.
1.2.3.4.1. Training with Fully Sharded Data Parallels (FSDP) CPU offloading
Red Hat Enterprise Linux AI version 1.2 now supports PyTorch’s Fully Sharded Data Parallels (FSDP) tool. You can now use FSDP during your training runs on RHEL AI.
1.2.3.4.2. Continuing training runs
Red Hat Enterprise Linux AI version 1.2 now allows you to continue a training run that may have failed during multi-phase training. You can continue a training run by running the ilab model train
command with the now supported --training-journal
flag that points to a YAML file that was generated during a prior multi-phase training run. This takes the training data that was already generated and continues training using that data. For more details on continuing training, see Continuing a training run.
1.2.3.5. Benchmark evaluation
Red Hat Enterprise Linux AI includes the ability to run benchmark evaluations on the newly trained models. On your trained model, you can evaluate how well the model knows the model you added with the MMLU_BRANCH
benchmark. For more details on benchmark evaluation, see Evaluating your new model.
1.3. Red Hat Enterprise Linux AI feature tracker
1.3.1. Installation feature tracker
Feature | 1.1 | 1.2 |
---|---|---|
Installing on bare metal | Generally available | Generally available |
Installing on AWS | Generally available | Generally available |
Installing on IBM Cloud | Generally available | Generally available |
Installing on Azure | Not available | Generally available |
Installing on GCP | Not available | Technology preview |
1.3.2. Platform support feature tracker
Feature | 1.1 | 1.2 |
---|---|---|
Bare metal | Generally available | Generally available |
AWS | Generally available | Generally available |
IBM Cloud | Not available | Generally available |
Azure | Not available | Generally available |
Google Cloud Platform | Not available | Technology preview |
Feature | 1.1 | 1.2 |
---|---|---|
Bare metal | Generally available | Generally available |
AWS | Generally available | Generally available |
IBM Cloud | Generally available | Generally available |
Azure | Not available | Generally available |
Google Cloud Platform (GCP) | Not available | Technology preview |
1.4. Technology preview feature status
1.4.1. Large Language Models (LLMs) technology preview status
Feature | 1.1 | 1.2 |
---|---|---|
| Generally available | Generally available |
| Generally available | Generally available |
| Technology preview | Technology preview |
| Technology preview | Technology preview |
| Generally available | Generally available |
| Generally available | Generally available |
1.5. Known Issues
1.5.1. Auto detecting a machine L40S accelorators
On Red Hat Enterprise Linux AI version 1.2, if you are using a machine with L40S accelerators, the CLI hardware auto detection displays that the L40S is the approxomite training profile for your system, even though the profile is an exact match. The L40S training profile is the appropriate training profile for this system.
1.5.2. The ilab model download
command does not show progress bar
The 1.2 version of RHEL AI does not show the progress of downloading models onto your system. This issue will be fixed in a later version of RHEL AI.
1.5.3. GUI AMD technology preview installations
Red Hat Enterprise Linux AI version 1.2 currently does not support graphical based installation with the technology previewed AMD ISOs. Ensure that the text
parameter in your kickstart
file is configured for non-interactive installs. You can also pass inst.text
in your shell during interactive installation to avoid an install time crash.
1.5.4. Kdump over nfs
Red Hat Enterprise Linux AI version 1.2 does not support kdump over nfs without configuration. To use this feature, run the following commands:
mkdir -p /var/lib/kdump/dracut.conf.d echo "dracutmodules=''" > /var/lib/kdump/dracut.conf.d/99-kdump.conf echo "omit_dracutmodules=''" >> /var/lib/kdump/dracut.conf.d/99-kdump.conf echo "dracut_args --confdir /var/lib/kdump/dracut.conf.d --install /usr/lib/passwd --install /usr/lib/group" >> /etc/kdump.conf systemctl restart kdump