IBM has unveiled the latest iteration of its Granite model, version 3.0. This update follows the initial release of Granite 1.0 back in May.
Granite is a large language model (LLM) that is open source under the permissive Apache 2.0 license. It boasts higher performance capabilities compared to other open source or commercial LLM models.
The new Granite 3.0 comes in two parameter sizes: 2B and 8B. IBM showcases the benchmark scores on the Hugging Face OpenLLM Leaderboard, demonstrating its ability to outperform similarly sized open source models like Llama 3.1 8B and Mistral 7B.
One standout feature of Granite 3.0 is its architecture design known as Granite Mixture of Experts (MoE), which enables fast response times (low latency) and reduces model running costs. The smaller 2B model can run on systems with just a single CPU.
Currently, Granite 3.0 is available for use on various cloud platforms such as IBM watsonx, Google Cloud Vertex AI, NVIDIA NIM, with the model downloadable from Hugging Face.
TLDR: IBM introduces Granite 3.0, a high-performing LLM model with updated features and architecture for enhanced efficiency and cost-effectiveness.
Leave a Comment