The UK-LLM initiative is developing an AI model based on NVIDIA Nemotron to reason in both English and Welsh, benefiting the 850,000 Welsh speakers. This new model aims to support public services like healthcare and education in Welsh, aligning with the Welsh government’s goal of a million Welsh speakers by 2050.

The UK-LLM project, led by University College London, collaborated with Bangor University and NVIDIA to develop the Welsh model. Welsh language experts from Bangor University are ensuring the model captures linguistic nuances, with plans to make the model and datasets available for enterprise and public sector use.

The model for Welsh is based on NVIDIA Nemotron and was trained on Welsh-language data using NVIDIA NIM microservices. The model development team utilized GPU clusters and NVIDIA GH200 Grace Hopper Superchips on Isambard-AI for translation and training workloads, supplementing existing Welsh data.

The framework used for the Welsh model development can serve as a foundation for multilingual AI development globally. Nemotron models are publicly available for developers to build reasoning models tailored to any language, domain, or workflow, offering cost-effective compute options from laptops to the cloud. Europe’s enterprises can run sovereign models on the Perplexity AI-powered search engine.

Read more at NVIDIA: Bringing AI to UK Languages With NVIDIA Nemotron