InstructLab
The open source project that lets the person who knows your business teach a model, without that person being a data scientist. For a shop whose knowledge lives in three people's heads, this is the interesting one.
Every AS/400 shop has the same quiet problem. The reason the month-end job has to run in that exact order lives in one person's head, and that person is three years from retiring. No manual covers it. Nobody has ever written it down.
InstructLab is an attempt to make writing it down actually worth doing. It is an open source project from IBM Research and Red Hat that lets someone who knows the subject, rather than someone who knows machine learning, add knowledge to a model.
How it works, without the jargon
Somebody writes down what they know
Not thousands of examples. A handful, in a structured folder arrangement InstructLab calls a taxonomy, organised as knowledge and skills.
The tool multiplies those examples
InstructLab generates a large set of synthetic training examples from the small set a human wrote. This is the step that used to require a research team.
The model gets tuned on the result
The output is your own version of an open model that knows your material, running on your own hardware, with nothing sent anywhere.
The underlying method is called LAB, which stands for Large-scale Alignment for chatBots, and it came out of IBM Research. Red Hat packages it inside Red Hat Enterprise Linux AI so it arrives already fitted to a Granite model.
Why this belongs in a hardware conversation
Because tuning is the one AI task where sending your data to somebody else's cloud is hardest to justify. Your operational knowledge is the asset. InstructLab plus an open Granite model plus a Power system you already own means the knowledge and the model both stay on your floor.
InstructLab tunes a model. It does not connect that model to Db2 for i, and it is not a shortcut around writing anything down. Somebody still has to sit with the person who knows and record what they know. The tool makes that effort go further, it does not replace it.
Related
Red Hat Enterprise Linux AI (RHEL AI)
A bootable Linux image with IBM Granite models and the InstructLab tuning tools already inside. It is the shortest route from bare metal to a working model, and it is where Granite and Red Hat stop being two separate decisions.
IBM Granite Models
IBM's own family of open language models, and the cargo an accelerator card is built to carry. You can download them free, run them on your own hardware, and nobody meters the tokens.
Red Hat AI Inference Server
One of the two products IBM names as required before a Spyre card will do anything. It is the driver, the traffic controller, and the reason a 128 GB card can serve more than one person at a time.
Red Hat OpenShift AI
The other product IBM accepts for driving a Spyre card. It is the heavier option: a full container platform for teams running several models, several projects, and more than one server.