3.2 good to miss

Hello there. Let's get you caught up on the recent launch of our Granite 3.2 models.👇

Granite 3.2, the latest release in our third generation of our Granite foundation models, is an essential step in the evolution of the Granite series beyond straightforward language models.

The release also includes an array of improvements to the efficiency, efficacy and versatility of our existing offerings. The models continue to deliver competitive performance at a fraction of the cost of state-of-the-art large models.

Here's What's New...

Reasoning when you need it

Granite 3.2 Instruct 8B and Granite 3.2 Instruct 2B, have been trained to offer enhanced reasoning capabilities relative to their 3.1 counterparts. Our implementation of reasoning runs somewhat counter to certain industry trends in order to maintain our practical approach to enhancing model performance.

🔷 The model’s internal reasoning process can be easily toggled on and off, ensuring the appropriate use of compute resources for the task at hand.

🔷 Our methodology brings the benefits of reasoning while preserving general performance and safety across the board.

🔷 You can tap into Granite 3.2’s thought process when it’s necessary or prioritize efficiency when it isn’t.

Granite goes multimodal

Granite Vision 3.2 2B is a lightweight large language model with computer vision capabilities that target everyday enterprise use cases, trained with a particular focus on visual document understanding. Handling both image and text inputs, Granite Vision 3.2's performance on essential document understanding benchmarks, such as DocVQA and ChartQA, rivals that of even significantly larger open models.

Slimmer, safer, more specific

Granite Guardian 3.2, the latest generation of our guardrail models designed to detect risks in prompts and responses, provides performance on par with Guardian 3.1 equivalents at greater speed with lower inference costs and memory usage by:

🔷 Introducing verbalized confidence, a new feature that provides a more nuanced evaluation of detected risks to acknowledge the ambiguity inherent to certain safety monitoring scenarios.

🔷 Incorporating two new model sizes: Granite Guardian 3.2 5B and Granite Guardian 3.2 3B-A800M, which maintain the performance of the larger models they were derived from.

👉 Granite 3.2: Get the key info now

Watch & Learn

Matt Upham walks you through how our Granite 3.2 models can generate code, debug errors, and even fix bugs from just a screenshot. Here's what you'll learn. ⤵️

• How to build a full-stack app with Next.js, React, TypeScript, Express, and Postgres.

• How Granite’s vision model can identify and explain errors from a screenshot.

• How the reasoning model suggests the right fix and improves debugging efficiency.

• Why our open-source AI models are game-changers for enterprise developers.

Kate Soule, Director, Technical Product Management, Granite, breaks down the release of Granite 3.2 in this episode of Mixture of Experts. Skip ahead to 20:11 to get right to it. ⤵️

👁️🐝Ⓜ️

Let us help you keep up with the speed of AI. Tune into our AI in Action Podcast to learn how to put AI into practice: ibm.biz/BdKFgN