IBM® Granite™ is a family of artificial intelligence (AI) models purpose-built for business, engineered from scratch to help ensure trust and scalability in AI-driven applications. Open source Granite models are available today.
IBM is named a strong performer in the The Forrester Wave™: AI Foundation Models for Language, Q2 2024.
Read IBM’s point-of-view on enterprise-grade models
With a principled approach to data transparency, model alignment and security red teaming, IBM has been delivering truly open source Granite models under an Apache 2.0 license to empower developers to bring trusted, safe generative AI into mission-critical applications and workflows.
IBM Granite models deliver best-in-class performance in coding, and above-par performance in targeted language tasks and use cases at lower latencies, with continuous, iterative improvements by using pioneering techniques from IBM Research® and contributions from open source.
With a fraction of the compute capacity, inferencing costs and energy consumption demanded by general-purpose models, Granite models enable developers to experiment, build and scale more generative AI applications while staying well within the budgetary limits of their departments.
IBM Granite code models (3B, 8B, 20B, 34B) trained on 116 programming languages demonstrate superior performances in code synthesis, fixing, explanation, editing and translation, across most major languages including Python, JavaScript, Java, Go, C++ and Rust.
When tuned with the InstructLab alignment technique developed by IBM Research, Granite language models (7b open source, 13B English, 20b multilingual, 8b Japanese) deliver accuracy and throughput on par with larger models at one-third the latency, and using only a fraction of GPU resources.
The time series and geospatial Granite models available on open source can be fine-tuned with your target data on your local computer, to deliver accurate forecasts and scientific computations faster, requiring only a fraction of compute power.
Lightweight, dependable and low-cost or no-cost models and tools to swiftly prototype their myriad ideas before they can scale them on production systems.
Models built on the foundation of transparent data sources, free from hate, abuse and profanity and unlicensed content, with targeted functions and trusted output.
Collaboration with thousands of developers who are already harnessing open source models to advance science, modernize code, improve productivity and transform experiences.
AI assistants are applications powered by Granite models and the IBM watsonx™ platform, deployed to automate workflows and implement AI across various technical and business functions.
Started originally in October 2023 with the first set of open source models released in May 2024, Granite has received recognition and validation from analysts, media and industry.
In testing against a range of other models, including those that have been opened under Apache 2.0 licenses, and more proprietary models, we found Granite models are competitive at a range of coding tasks.
A new report from Stanford University’s Center for Research on foundation models showed that IBM’s model scored a perfect 100% in several categories designed to measure how open models are.
According to Forrester, the Granite family of models provides enterprise users with some of the most robust and clear insights into the underlying training data. This is important for efficiently refining model behavior for specific use cases and domains, and for protecting enterprises from risk due to any unlicensed content in the training data.
IBM believes in the creation, deployment and utilization of AI models that advance innovation across the enterprise responsibly. IBM watsonx AI and data platform have an end-to-end process for building and testing foundation models and generative AI. For IBM-developed models, we search for and remove duplication, and we employ URL blocklists, filters for objectionable content and document quality, sentence splitting and tokenization techniques, all before model training.
During the data training process, we work to prevent misalignments in the model outputs and use supervised fine-tuning to enable better instruction following so that the model can be used to complete enterprise tasks via prompt engineering. We are continuing to develop the Granite models in several directions, including other modalities, industry-specific content and more data annotations for training, while also deploying regular, ongoing data protection safeguards for IBM developed models.
Given the rapidly changing generative AI technology landscape, our end-to-end processes are expected to continuously evolve and improve. As a testament to the rigor IBM puts into the development and testing of its foundation models, the company provides its standard contractual intellectual property indemnification for IBM-developed models, similar to those it provides for IBM hardware and software products.
Moreover, contrary to some other providers of large language models and consistent with the IBM standard approach on indemnification, IBM does not require its customers to indemnify IBM for a customer's use of IBM-developed models. Also, consistent with the IBM approach to its indemnification obligation, IBM does not cap its indemnification liability for the IBM-developed models.
The current watsonx models now under these protections include:
(1) Slate family of encoder-only models.
(2) Granite family of a decoder-only model.