NVIDIA Expands AI Training with Multilingual Workshop at GTC Paris

NVIDIA is enhancing its AI training offerings with a new multilingual workshop at the upcoming GTC Paris event, according to a blog post by the company. This initiative by the NVIDIA Deep Learning Institute aims to equip developers with the skills necessary to transform open-source large language models (LLMs) into domain-specialized and genuinely multilingual AI assets.

Addressing Multilingual Challenges

Despite the capabilities of LLMs in recognizing and generating content, they often fall short when dealing with specialized business knowledge or diverse linguistic and cultural contexts. Many models, labeled as multilingual, are predominantly trained in English, which can lead to inaccuracies and cultural insensitivity. NVIDIA’s new workshop, “Adding New Knowledge to LLMs,” will tackle these issues by providing hands-on training in model adaptation to improve accuracy across various languages and sectors.

Importance of Multilingual Model Evaluation

Evaluating models in multiple languages is crucial for their development and customization. The workshop emphasizes the significance of multilingual evaluation, as many Europeans, for example, prefer using their native languages over English. Models like Llama 2 are trained on less than 5% non-English data, highlighting the need for rigorous multilingual evaluation to avoid misleading performance metrics and costly deployment errors.

Workshop Overview and Key Tasks

The “Adding New Knowledge to LLMs” workshop is a full-day, instructor-led session that will guide participants through the lifecycle of AI model customization. Key tasks include systematic evaluation and dataset creation, advanced data curation, targeted knowledge injection, and model optimization for domain and language. Participants will learn to use tools like NVIDIA NeMo Evaluator and NeMo Model Optimizer to enhance their models’ multilingual capabilities and domain-specific performance.

Real-World Applications and Impact

NVIDIA’s collaborative efforts with global organizations have already shown significant improvements in multilingual AI. Partnerships with entities like the Barcelona Supercomputing Centre and EuroLLM have resulted in improved language-specific task accuracy and the development of powerful multilingual models. These advancements underscore the importance of NVIDIA’s initiatives in advancing multilingual AI capabilities.

For those interested in exploring the latest AI innovations, the GTC Paris event offers an opportunity to participate in this workshop and other related sessions. More information about the event and registration can be found on the NVIDIA blog.

Image source: Shutterstock