Here's the key info, at a glance:
- The new Granite 3.2 8B Instruct and Granite 3.2 2B Instruct offer experimental chain-of-thought reasoning capabilities that significantly improve their ability to follow complex instructions with no sacrifice to general performance. The reasoning process can be toggled on and off, allowing for efficient use of computing resources.
- When combined with IBM’s inference scaling techniques, Granite 3.2 8B Instruct’s extended thought process enables it to meet or exceed the reasoning performance of much larger models, including GPT-4o and Claude 3.5 Sonnet.
- Our new multimodal model, Granite Vision 3.2 2B, was developed with a particular focus on document understanding, on which it matches the performance prominent open models 5 times its size.
- The latest additions to the Granite Timeseries model family, Granite-Timeseries-TTM-R2.1, expand TTM’s forecasting capabilities to include daily and weekly predictions in addition to the minutely and hourly forecasting tasks already supported by prior TTM models.
- We’re introducing new model sizes for Granite Guardian 3.2, including a variant derived from our 3B-A800M mixture of experts (MoE) language model. The new models offer increased efficiency with minimal loss in performance.
- The Granite Embedding model series now includes the ability to learn sparse embeddings. Granite-Embedding-30M-Sparse balances efficiency and scalability across diverse resource and latency budgets.
- Like their predecessors, all new IBM Granite models are released open sourced under a permissive Apache 2.0 license.
- Granite 3.2 models are now available on IBM watsonx.ai, Hugging Face, Ollama, LMStudio, and Replicate.
Granite 3.2, the latest release in our third generation of IBM Granite models, is an essential step in the evolution of the Granite series beyond straightforward language models. Headlined by experimental reasoning features and our first official vision language model (VLM), Granite 3.2 introduces several significant new capabilities to the Granite family.
The release also includes an array of improvements to the efficiency, efficacy and versatility of our existing offerings. IBM’s prioritization of practical, enterprise-ready models continues the pursuit of state-of-the-art performance with fewer and fewer parameters.
The latest Granite models are open sourced under Apache 2.0 license. All Granite models are now available on Hugging Face. Select models are also available through additional platforms, including IBM watsonx.ai. Tutorials, recipes and other resources are available in the "Getting Started" section at the end of this article.
#GenerativeAI