IBM, a global technology leader, has officially introduced the first models in the Watsonx Granite model series, revolutionizing generative AI for seamless integration into various business applications and workflows. The Granite models mark a significant step forward in leveraging generative AI across language and code modalities, empowering businesses to scale AI effectively.
The Granite models are designed in various sizes, all based on a decoder-only architecture, aimed at aiding businesses in scaling their AI capabilities. Notable applications include retrieval augmented generation to tailor responses from enterprise knowledge bases, summarization to condense lengthy content like contracts or call transcripts, and insight extraction and classification to analyze customer sentiment.
An essential aspect of IBM’s approach is enabling businesses to be AI value creators. They can bring their proprietary data to IBM base models and build unique models tailored to their specific business and use cases. IBM’s investment in developing specialized foundation models is crucial, ensuring precision and accuracy tailored to the focused requirements of business use cases.
IBM’s dedication to flexibility is evident through its support for third-party models such as Meta’s Llama 2-chat 70 billion parameter model and those from the Hugging Face community, offering a broader range of options for businesses.
In a commitment to transparency and responsible AI deployment, IBM has published information about the training methodology for Granite models. These foundation models have been meticulously trained on business-relevant datasets from various domains, including internet, academic, code, legal, and finance. The training data underwent rigorous filtration to eliminate objectionable content, benchmarking against internal and external models to ensure responsible deployment and mitigate key concerns such as governance, risk assessment, privacy, and bias.
Furthermore, IBM’s standard intellectual property protection, akin to hardware and software products, applies to the IBM-developed Watsonx models. This contractual protection, in the form of IP indemnity, assures clients of the responsible use of AI models while providing the freedom to use their data and develop AI applications while benefiting from the accuracy and trust offered by IBM foundation models.
Dinesh Nirmal, Senior Vice President, Products, IBM Software, emphasized the importance of AI technologies that showcase success at scale and encompass responsible use practices and guardrails. The release of the IBM Granite model series underscores IBM’s dedication to providing businesses with cutting-edge AI suited to their unique business needs, emphasizing a robust end-to-end model lifecycle management process within the Watsonx AI and data platform.