IBM has announced the release of the first models in the watsonx Granite series, which utilize generative AI for language and code tasks. 

These models come in various sizes to cater to different business needs and are based on a decoder-only architecture. They can be used to scale AI in various ways, such as generating customized responses from enterprise knowledge bases, summarizing lengthy content like contracts or call transcripts, and extracting insights and classifying data, including customer sentiment.

In addition to its own models, IBM provides flexibility to use third-party models like Meta’s Llama 2-chat 70 billion parameter model and models from the Hugging Face community.

“When it comes to today’s AI innovation boom, the businesses that are positioned for success are the ones outfitted with AI technologies that demonstrate success at scale and have built-in guardrails and practices that enable their responsible use,” said Dinesh Nirmal, senior vice president of products at IBM Software. “Today’s release of IBM’s Granite model series and commitment to stand behind IBM-developed watsonx models is a testament to IBM’s end-to-end model lifecycle management process in its watsonx AI and data platform that delivers businesses cutting-edge AI outfitted for their unique business needs.”

IBM has developed foundational AI models that have been trained using datasets from five different domains: internet, academic, code, legal, and finance. These models have been carefully curated for business applications by IBM. The training data underwent a rigorous filtering process to remove objectionable content and was benchmarked against both internal and external models. 

According to IBM, this effort aims to ensure responsible deployment of AI and address issues like governance, risk assessment, privacy, and bias mitigation. IBM is utilizing its AI and data model lifecycle governance process to manage and mitigate client risk with the watsonx AI and data platform. Additionally, they plan to release watsonx.governance, an AI governance toolkit, later this year to facilitate trusted AI workflows.

IBM has also confirmed that its standard intellectual property protections will apply to these AI models.