IBM Launches 'Granite' AI Models for Businesses
The Granite models were developed specifically for business, not consumer, use, and have been trained on datasets from five areas.
Big Blue recently announced the general availability of AI foundation models that have been tailored for business use cases such as customer service and customer sentiment analysis.
IBM's Granite foundation models, available via the company's watsonx AI platform, let companies take advantage of generative AI functionality in their day-to-day processes. Per IBM's announcement:
[B]usinesses can use them to apply retrieval augmented generation for searching enterprise knowledge bases to generate tailored responses to customer inquiries; use summarization to condense long-form content -- like contracts or call transcripts -- into short descriptions; and deploy insight extraction and classification to determine factors like customer sentiment.
The Granite models were developed specifically for business -- not consumer -- use, and have been trained on datasets from these five areas: academic, code, financial, internet and legal. To ensure "responsible deployment" and mitigate issues around bias, privacy and governance, the Granite training data was "filtered for objectionable content and benchmarked against internal and external models."
The result, according to IBM, is a generative AI framework that's safe enough for businesses to "bring their proprietary data to IBM base models and build a model that is unique to their business and use cases."
Currently, there are two types of Granite models available: granite-13b-chat and granite-13b-instruct. Both are 13-billion parameter models, which, per IBM, potentially makes them more efficient and cost-effective than larger models.
Granite is one of two foundation model families currently available on watsonx. The other, dubbed Slate, are for non-generative AI use cases.
More information on IBM's watsonx platform is available here.