Introducing support for the IBM Granite 3.0 Model family.

[kailas711]

IBM's Granite 3.0 series offers a range of models with both Mixture of Experts (MoE) and dense architectures in sizes of 2B and 8B parameters.

Link: https://huggingface.co/ibm-granite

Integrating this series enhances LLM-Vec capabilities, enabling both MoE and dense configurations within a single model family Granite. It’d be great to have these models added to the library for large-scale NLP tasks.