Skip to content

IBM Unveils Granite 4: New Open-Source Language Models Boost Efficiency

IBM's Granite 4 models offer a mix of Transformer and Mamba architectures for better performance. They're now accessible on IBM's platforms and coming soon to Amazon and Microsoft services.

In this image there is a table with many cores, a laptop, a pen and a few things on it.
In this image there is a table with many cores, a laptop, a pen and a few things on it.

IBM Unveils Granite 4: New Open-Source Language Models Boost Efficiency

IBM has unveiled Granite 4, the latest generation of open-source language models. This series combines two neural network architectures, Transformer and Mamba, and offers improved efficiency over previous generations.

Granite 4 models range from 3 to 32 billion parameters, with the largest, Granite-4.0-H-Small, utilising a mixture-of-experts design for enhanced customer support. The series is built on the latest version of Mamba, Mamba 2, which reduces memory pressure, particularly with long input prompts.

IBM plans to offer these models through various platforms. Currently, they are available on IBM's watsonx.ai platform and external services like Hugging Face. In the future, IBM aims to make them accessible through Amazon SageMaker JumpStart and Microsoft Azure AI.

Granite 4 models, utilising Transformer and Mamba architectures, provide improved performance and efficiency. They are currently available through IBM's platforms and will soon be accessible via Amazon and Microsoft services, expanding their reach and usability.

Read also:

Latest