IBM has announced the availability to enterprises of the first series of generative AI models from the watsonx Granite series, created to facilitate the integration of generative AI into applications and workflows. These models apply generative AI to language and code, and will have standard contractual measures to protect the intellectual property of IBM products.
These models are developed based on a decoder-only architecture, and seek to make it easier for companies to scale Artificial Intelligence. They can feed their data into IBM’s foundational models and generate a unique model for your use cases and your business.
IBM’s foundational models are tailored to meet the specific requirements of enterprise use cases, which are much more unique than those of consumer AI, and therefore require very high levels of precision and accuracy. In addition to its own models, the company offers flexibility to use third-party models, such as the Llama 2 from Meta or those from the Hugging Face community.
To train these models developed by IBM, data sets relevant to companies from five environments have been used: Internet, academic, code, legal and financial). The data used for training was filtered to locate objectionable content, and compared with internal and external models. In this way, the company sought to facilitate responsible deployment, addressing issues such as governance, risk assessment, privacy issues, and bias mitigation.
IBM also plans to launch watsonx.governance, a suite of AI governance tools focused on enabling trusted Artificial Intelligence workflows, before the end of this year. This is because the company is using its AI model and data and capabilities lifecycle governance procedure. This reduces customers’ risk when using Watsonx AI and the IBM data platform.
Dinesh Nirmal, Vice President of Products, IBM Softwarehas stressed that «In the current boom in AI innovation, companies positioned for success are those that are equipped with AI technologies that demonstrate success at scale and have built-in barriers and procedures that enable their responsible use. The launch of IBM’s Granite model series and commitment to supporting IBM-developed watsonx models is a testament to IBM’s end-to-end model lifecycle management process on its watsonx data and AI platform, which offers cutting-edge AI tailored to your unique business needs«.