Chapter 1. Overview of enabling LAB-tuning

Data scientists can use LAB-tuning in OpenShift AI to run an end-to-end workflow for customizing large language models (LLMs). The LAB (Large-scale Alignment for chatBots) method provides a more efficient alternative to traditional fine-tuning by using taxonomy-guided synthetic data generation (SDG) combined with a multi-phase training process. LAB-tuning workflows can be launched directly from the OpenShift AI dashboard using the preconfigured InstructLab pipeline, simplifying the tuning process.

LAB-tuning depends on several OpenShift AI components working together to support model customization. It uses data science pipelines to run the tuning workflow, KServe to deploy and serve the teacher and judge models, and the Training Operator to run distributed model training across GPU-enabled nodes. LAB-tuning also relies on the model registry to manage model versions, storage connections (such as S3 or OCI) to store pipeline artifacts and model outputs, and GPU hardware profiles to schedule training workloads.

To enable LAB-tuning, an OpenShift cluster administrator must configure the required infrastructure and platform components by completing the following tasks:

A cluster administrator or an OpenShift AI administrator must perform additional setup within the OpenShift AI dashboard: