IBM has unveiled its latest family of AI models called Granite 3.0, at its annual TechXchange event. It said that these models can outperform or match similarly sized models on academic and industry benchmarks. Unlike OpenAI which is secretive about its models, IBM has released the Granite models under the permissive Apache 2.0 license.
The new Granite 3.0 family includes lots of new models. They are:
- General Purpose/Language: Granite 3.0 8B Instruct, Granite 3.0 2B Instruct, Granite 3.0 8B Base, Granite 3.0 2B Base
- Guardrails & Safety: Granite Guardian 3.0 8B, Granite Guardian 3.0 2B
- Mixture-of-Experts: Granite 3.0 3B-A800M Instruct, Granite 3.0 1B-A400M Instruct, Granite 3.0 3B-A800M Base, Granite 3.0 1B-A400M Base
The company said that the Granite 3.0 8B and 2B models are ideal workhorse models for enterprise AI, meaning they're more robust and reliable and can handle a wider set of tasks. IBM said these models offer strong performance for tasks like Retrieval Augmented Generation (RAG), classification, summarization, entity extraction, and tool use. When deployed, these models can be fine-tuned with enterprise data to integrate across diverse business environments and workflows.
IBM's new Granite models are now available for download on HuggingFace under the Apache 2.0 license. The Granite Guardian 3.0 8B and 2B models are also available for commercial use via IBM's watsonx platform and a selection of Granite 3.0 models will also come to Nvidia NIM microservices and Google Cloud's Vertex AI Model Garden.
If you are a programmer and want to get tasks done quicker, Granite's code capabilities can also be utilized through the Visual Studio Code extension, IBM Granite.Code. This extension can help you in lots of programming languages including C, C++, Go, Java, and Python.
A final note of interest with the Granite 3.0 family is that they lay the groundwork for IBM's agentic AI efforts. The Granite 3.0 8B model features advanced reasoning and a highly structured chat template and prompting style for implementing tool-use workflows. We will see more from the company with regards to agentic AI next year, it said, including pre-built agents for specific domains and use-cases. Agentic AI, for those who don't know, is capable of acting more independently to complete tasks.
2 Comments - Add comment