IBM Releases Watsonx Granite Models
Written by Kay Ewbank   
Tuesday, 10 October 2023

IBM has released the first models in its watsonx Granite model series. Watsonx is an AI and data platform that includes an AI development studio with access to IBM-curated and trained foundation models and open-source models, access to a data store, and a toolkit for AI governance.

IBM Watson is a platform that uses natural language processing and machine learning to discover insights in large sets of unstructured data. It is a question-answering system that can answer questions phrased using natural language.

watsonlogo

The initial release of watsonx.ai included the Slate family of encoder-only models targeting enterprise Natural Language Processing (NLP) use. This latest release is aimed at uses such as summarization, content generation and extracting insights.

The Granite series is a collection of generative AI models that apply generative AI to the modalities of language and code. IBM says the Granite models are being developed in different sizes and are built on a decoder-only architecture. The models are trained on business-relevant domains, including financial, legal, IT, coding, and academic domains. The model type lends itself to retrieval augmented generation for searching knowledge bases to generate tailored responses to customer inquiries, or to condense content into short descriptions.

The training data was filtered for objectionable content and benchmarked against internal and external models to address issues including governance, risk assessment, privacy concerns and bias mitigation. The Watson team says that as part of this process, IBM is using its end-to-end AI and data model lifecycle governance process and capabilities to manage and help mitigate client risk with the watsonx AI and data platform. This includes the planned release of watsonx.governance, an AI governance toolkit to enable trusted AI workflows that will be generally available later this year.

The initial Granite collection includes two variants of a 13 billion parameter model: granite-13b-chat and granite-13b-instruct. According to the Watson team, at 13 billion parameters, these variants have the potential to be more efficient than larger models since they can fit onto a single A100-80 GB GPU which can also help lower the total cost of ownership.

IBM says its standard intellectual property protections will apply to the AI models.

The Watsonx Granite data sets are available now.

watsonlogo

More Information

Watsonx Website

Related Articles

IBM Announces WatsonX AI Platform

IBM Announces COBOL Modernizing Solution

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Banner


GR00T Could Be The Robot You Have Always Wanted
27/03/2024

We may not have flying cars, but we could well soon have robots that match up to predictions for the 21st century. Nvidia has announced GR00T, a cleverly named project to build robots using foundation [ ... ]



Rust Twice As Productive As C++
03/04/2024

Google director of engineering, Lars Bergstrom, gave a talk at the recent Rust Nation UK conference and claimed that Rust was twice as productive as C++. Given how good Google is at C++, this is quite [ ... ]


More News

raspberry pi books

 

Comments




or email your comment to: comments@i-programmer.info

Last Updated ( Tuesday, 10 October 2023 )