Introducing Granite 3.2; Next-Gen Multi-Modal And Reasoning AI
IBM introduced Granite 3.2. This launch comes as a continued effort to deliver efficient yet compact and practical enterprise AI for real-world impact.
IBM introduced Granite 3.2 | Official website
With the constant advancement of artificial intelligence (AI), there has been a need for better large language models (LLMs). To contribute in this, IBM is all set expand the next generation of its Granite large language model (LLM) family.
IBM introduced Granite 3.2. This launch comes as a continued effort to deliver efficient yet compact and practical enterprise AI for real-world impact. Here's a detailed look at what the new LLMs can bring to the table.
Decoding Granite 3.2: What's In
The new Granite 3.2 8B Instruct and Granite 3.2 2B Instruct can deliver experimental chain-of-thought reasoning capabilities. This is aimed at significantly improving their ability to follow complex instructions. Plus there will be no sacrifice to general performance. The reasoning process can be toggled on and off. Thus, allowing for efficient use of computing resources. Moreover, Granite Vision 3.2 2B, was developed with a specific focus on document understanding,
The new model sizes for Granite Guardian 3.2, includes a variant similar to 3B-A800M mixture of experts (MoE) language model. The new models also offer high-end efficiency with minimal loss in performance.
Granite Guardian 3.2: Key Highlights
So, what makes Gaurdian 3.2 different from its earlier models? Given below are key features:
The Granite Embedding model series now includes the ability to learn sparse embeddings.
Granite-Embedding-30M-Sparse is made to balance efficiency and scalability across diverse resource and latency budgets.
Granite 3.2 is an essential step in the evolution of the Granite series beyond straightforward language models.
The model is then “healed” by retraining it on 80% of the original training data, after which 2 more layers are pruned,
To conclude, all Granite 3.2 models are available under the permissive Apache 2.0 license on Hugging Face. You can get selected models available on Ollama, IBM watsonx.ai, Replicate, and LM Studio. It is expected to be available in RHEL AI 1.5 also soon.
RECENT STORIES
-
-
-
-
-