IBM announced today the launch of Granite 3.2, the next generation of its large language model (LLM) product line, on the 27th. Granite is a solution that embodies IBM's ongoing efforts to provide small, efficient, and practical artificial intelligence (AI) for corporations to generate tangible business impact.
The Granite 3.2 model is available under the Apache 2.0 license permitted by Hugging Face. Some models will be available immediately today on IBM watsonx.ai, Ollama, Replicate, and LM Studio, and soon it will be provided on Red Hat Enterprise Linux (RHEL) AI 1.5, offering new features for corporations and the open-source community.
IBM introduced a feature to Granite AI that allows for the programming of 'chain reasoning' to be activated or deactivated. For simple tasks, this model operates without inference, thereby reducing unnecessary computational load. Additionally, the Granite 8B model has been shown to perform at least as well as, or better than, larger models on standard mathematical reasoning benchmarks through other inference techniques such as inference expansion.
IBM implements the next-generation time series model, TinyTimeMixers (TTM), capable of long-term forecasts up to two years into the future, with Granite 3.2 instruct, vision, and guardrail models, which have fewer than 10 million parameters.
Sriram Raghavan, vice president of IBM AI Research, noted, 'The next AI era will be characterized by efficiency, integrability, and practicality, enabling corporations to achieve strong performance without excessive computing expense.' He added, 'I believe IBM's latest Granite model, focused on open solutions, has helped enhance the accessibility, cost-effectiveness, and value of AI for today's corporations.'