What is an open source model?

The largest problem is that AI models are binaries, and the "source code" isn't just the data, it's so much more.

tl;dr: Everyone is debating what "open source" means and we can't seem to come to an agreement, but the Model Openness Framework seems to be the most supported.

What is InstructLab and what does it solve?

What is fine tuning?

tl;dr: InstructLab is an opinionated workflow for fine tuning models.

A qna.yaml example

A source markdown document example

RAG vs Fine Tuning

Imagine an AI model as a Librarian

RAG: Retrieval-Augmented Generation or the "library"

Fine Tuning goes into the "brain", so sending the librarian "back to school"

Agents and Agentic Workflows

Let's talk quickly about agentic frameworks...

What is an "agent?"

A few major frameworks people are gravitating towards:

1. Bee Framework
2. CrewAI (a great crash course video)
3. AutoGen

But what model should we tune?

This screenshot was taken March 10, 2025... 1.4 million seems a lot.

And this brings us back to Granite

• Granite 3.2
• Granite Vision
• Granite Guardian
• Granite Embedding
• Granite Time Series
• Granite Code

Use Cases for Granite

ai-abstract-classifier: Run abstracts through a local LLM to give you some possible scores on what you are looking for.

Help teach Medical Students Motiviational Interviewing.

Leverage Granite 3.2 like this to create stories for some kids. This is a picture of me teaching AI at a STEM Night at an school.

I am figuring out the process of using Granite Time Series model to project my insulin pump supplies and insulin usage

And finally I'm using Granite Time Series model to project my disk and load usage with Nagios!

But JJ, how can I play with Granite?

Open Source Workshop

Spend some time learning how to get Granite and what is required to run it locally, securely, and privately on your laptop.

Granite Workshop

Learn how to leverage the different Granite models to integrate with applications and developer focused work.