-
Red Hat is open sourcing its InstructLab project for large language models
-
It thinks open source is the best way to get more developers involved in GenAI
-
Red Hat's CEO said GenAI models so far are "literally trained on all the information in the world," and that's too much for a lot of use cases
RED HAT SUMMIT DENVER — Software conferences are always slightly hilarious because they kick off with some great music and lights (see picture above) at 8 o’clock in the morning. Then a guy in a sweatshirt comes out and starts talking about software.
At the kickoff of the Red Hat Summit today, the audience was jazzed to hear about several generative artificial intelligence (GenAI) initiatives.
First, Red Hat said it is going to open source its InstructLab project for enhancing large language models (LLMs).
The InstructLab open source project will provide a cost-effective way for enterprises to align LLMs with data for their specific use cases. And it will open the door for people with minimal GenAI experience to get involved.
There are already a couple of open source GenAI projects, namely Meta’s Llama and Mistral. Now, InstructLab can be added to that list.
Speaking today at the summit, Red Hat CEO Matt Hicks said open source GenAI projects are important “because to unlock the real potential of AI you have to be able to close the gap in that last mile of knowledge for your use case.”
He said that the GenAI models to date are “literally trained on all the information in the world.” But that’s too much for most enterprises, which want to train models on their own specific data.
He gave the example of a financial firm that wants to train data from quarterly earnings reports in order to find anomalies. Hicks thinks that InstructLab will be great for training smaller models relevant for specific use cases.
Even though InstructLab is open source, developers will be able to keep their data private and yet still make upstream contributions to benefit the project.
Hicks concluded, “I feel fortunate to experience the convergence of AI and open source.”
RHEL AI
Another piece of news at today’s summit, which got a big applause from the audience, was that Red Hat announced the launch of Red Hat Enterprise Linux AI (RHEL AI).
RHEL AI includes the open source-licensed Granite LLM family from IBM Research.
Last year, Red Hat announced its OpenShift AI project. OpenShift is Red Hat’s bread and butter technology, which uses containers and the container management system Kubernetes as the foundation for cloud-native computing. And OpenShift AI combined the company's expertise in containers along with AI.
Today, the company said that the combination of OpenShift AI, plus RHEL AI, plus the Granite language and code models, along with the InstructLab project will help companies tune models for their specific use cases.
Ashesh Badani, Red Hat’s chief product officer, said, “The open source of the Granite language and InstructLab gives big results from smaller models.”
Podman AI Lab
Red Hat made an insane number of announcements today, but another one rising to the top was the announcement of Podman AI Lab. It gives developers the ability to build GenAI applications in containers, using their own laptops.
Several Red Hat developers on stage today said it was a big deal to be able to do this work on their own laptops. It saved a lot of time and budget. They also really liked that they could work with plain text.
Podman AI Lab features a recipe catalog with sample applications that give developers a jumpstart on some of the more common use cases for LLMs, including chatbots, text summarizers and code generators.
Summing it up
Justin Boitano, VP of Enterprise Products with NVIDIA, said today that cloud-native computing and containers have really paved the way for GenAI. He said when a user asks a GenAI platform a query, the platform creates a reasoning plan. It might call smaller models for information. These models are chained together as microservices running on OpenShift.
Finally, Intel’s CEO Pat Gelsinger made an appearance at today’s Red Hat Summit (albeit remotely). Gelsinger said, “I predict every company will become an AI company.” But he said their biggest hurdle will be unlocking the value of their data in a secure fashion.