International Business Machines Corporation or IBM is a technology company based in America and is the largest industrial research organization in the world. This company aimed at AI models for business intelligence and it has recently launched Granite 3.0. The aim of releasing such a model is to support more enterprises like custom service, automation in IT, cybersecurity, BPO or Business Process Outsourcing, and application development.
One of IBM’s latest launches is the third-generation AI model. It is a compact, fit-for-purpose, and open-sourced model that gives great performance across a wide range of enterprise tasks in cybersecurity. This approach differs from rivals such as Microsoft which charge customers for access to their models. To cover up for it, IBM offers a paid tool that helps run models inside data centers after they have been customized. This tool is called Wastonx.
The Granite 3.0 language model is a basic and instruction-tuned language model designed for agentic workflows. This model also summarizes text, analyses it, and extracts important information, it also generates content and classification can also be done here. The family of Granite 3.0 is specifically designed for business applications.
The decoder-only models are designed to generate code, explain the code, and edit them. It is trained with 116 programming languages. This model is lightweight and trained to run efficiently through many hardware configurations. The model also ensures data security and mitigating tasks with a variety of user prompts and LLM responses.
The new Granite 3.0 8B and 2B models are released under the permissive Apache 2.0 license showing strong performance across many academic and enterprise benchmarks which in turn, are able to outperform or match similar-sized models. Granite 3.0 builds responsible AI with hard detection capabilities, transparency, and IP protection.
The other features in Granite 3.0 are as follows:
- There are a mixture of experts like Granite 3.0 3B-A800 Instruct, Granite 3.0 1B-A400M Instruct, Granite 3.0 3B-A800 Base, and Granite 3.0 1B-A400M Base.
- Guardrails and Satey features are available. They include Granite 3.0 8B and Granite Guardian 3.0 2B.
- The general purpose/language is Granite 3.0 8B.
The Granite 3.0 8B and 2B language models are designed as workhouse models for enterprise AI; this delivers strong performance for tasks such as Retrieval Augmented Generation(RAG), tool use, and entity extraction. These compact models are designed to be fine-tuned with enterprise data and seamlessly integrated across diverse business environments. Vhgfhugt
The benchmarks set by Granite 3.0 are commendable. The Granite 3.0 8B Instruct model’s performance leads on average against the similar-sized open-source models from Meta and Mistral. On IBM’s state-of-the-art AttaQ safety benchmark, the Granite 3.0 8B Instruct model leads across all measured safety dimensions compared to other models from Meta and Mistral.2
Granite 3.0 models were trained on more than 12 trillion tokens, utilizing data from 12 distinct natural languages and 116 programming languages. This was done by a novel two-staged training method from several thousand experiments designed to optimize data quality and other parameters.
By the end of the year, the 3.0 8B and 2B language models are set to support an expanded 128K context window and multi-modal documents. Additionally, IBM will release updated pre-trained Granite Time Series models, following the initial versions launched earlier this year. These new models have been trained on three times the data and demonstrate strong performance across all major time series benchmarks, surpassing models from Google, Alibaba, and others that are ten times larger.
Keep Reading: Huawei’s Ascend 910C AI Chips Challenge Nvidia In China Market
