IBM has introduced the latest iteration of its Granite large language model (LLM) family, known as Granite 3.2. This release is part of IBM's ongoing efforts to provide small, efficient AI models for enterprise applications. The new models are available under the Apache 2.0 license on platforms like Hugging Face, with select models accessible on IBM watsonx.ai, Ollama, Replicate, and LM Studio.
The Granite 3.2 family includes a new vision language model (VLM) designed for document understanding tasks. It demonstrates performance comparable to larger models such as Llama 3.2 11B and Pixtral 12B on benchmarks like DocVQA and ChartQA. IBM utilized its open-source Docling toolkit to process millions of PDFs and generate synthetic question-answer pairs to enhance the VLM's capabilities.
Granite 3.2 also features enhanced reasoning capabilities in its 2B and 8B models through "chain of thought" functionalities that can be toggled on or off for efficiency optimization. This allows the 8B model to achieve significant improvements in instruction-following benchmarks without compromising safety or performance.
The size of Granite Guardian safety models has been reduced by 30% while maintaining their performance levels from previous versions. A new feature called verbalized confidence offers nuanced risk assessment in safety monitoring.
David Tan, CTO at CrushBank, commented on the advancements: "At CrushBank, we've seen first-hand how IBM's open, efficient AI models deliver real value for enterprise AI – offering the right balance of performance, cost-effectiveness, and scalability," adding that they are eager to explore the new reasoning capabilities in building agentic solutions.
Sriram Raghavan, VP of IBM AI Research, highlighted the importance of efficiency in AI development: "The next era of AI is about efficiency, integration, and real-world impact – where enterprises can achieve powerful outcomes without excessive spend on compute."
Additionally, IBM is releasing an updated version of its TinyTimeMixers (TTM) models with capabilities for long-term forecasting up to two years ahead. These are intended for use in trend analysis across various sectors including finance and retail.
To gain further insights into Granite 3.2's features and applications, readers are encouraged to consult a detailed technical article provided by IBM.
