UK-LLM Initiative Launches Welsh AI Model Using NVIDIA Nemotron Technology
Executive Summary
The UK-LLM initiative, a collaboration between University College London, NVIDIA, and Bangor University, has developed a new AI model capable of reasoning in both English and Welsh. Built using NVIDIA's open-source Nemotron framework and trained on the Isambard-AI supercomputer, the model aims to support Welsh-language public services like healthcare and education. This project represents a significant step in "sovereign AI," creating a blueprint for developing AI models for other minority languages in the UK and globally.
Key Takeaways
* Product/Service/Initiative Name: UK-LLM.
* Primary Function: An AI language model that can reason in Welsh to support the delivery of public services and preserve the language.
* Key Features & Capabilities:
* Based on NVIDIA's open-source Nemotron models (49B and 9B parameter versions).
* Trained on the UK's Isambard-AI supercomputer, which uses NVIDIA GH200 Grace Hopper Superchips.
* Overcame data scarcity by translating over 30 million entries from English to Welsh using NVIDIA NIM microservices to create a new training dataset.
* Bangor University provided linguistic expertise to ensure the model handles the nuances of the Welsh language accurately.
* The methodology is intended to be a template for creating models for other languages like Cornish, Irish, and Scottish Gaelic.
* Target Audience: Developers, enterprise users, and the public sector, particularly institutions in Wales.
* Availability: The model will be made available to developers through an API from AI cloud provider Nscale. The model and datasets are expected to be released for enterprise and public sector use.
* Stated Goal: To ensure Welsh remains a living language, help the Welsh government achieve its goal of one million speakers by 2050 ("Cymraeg 2050"), and make public services more accessible.
Strategic Importance
This announcement showcases a national "sovereign AI" strategy, leveraging public-private partnerships and national supercomputing infrastructure to create culturally specific AI tools. It establishes a repeatable framework for preserving and promoting minority languages through AI, positioning the UK and NVIDIA as leaders in this specialized domain.