NVIDIA releases French and Belgian persona datasets
NVIDIA has released synthetic persona datasets for France and Belgium, along with a fine-tuning walkthrough, so developers can adapt open Nemotron models to local rules as well as local languages.
NVIDIA published two new datasets on its technical blog: Nemotron-Personas-France, built with French AI startup Pleias, and Nemotron-Personas-Belgium, built with Pleias and KU Leuven. Both belong to NVIDIA's open Nemotron data program, which now has locale-grounded persona datasets for 12 countries.
Each persona is a synthetic resident. Its age, commune, job and household follow regional demographic, geographic, linguistic and cultural distributions, such as French census data, and it comes with written profiles of the person's work and daily life. The Belgian dataset covers Dutch, French, German and English. NVIDIA says these details change both how a user asks a question and what the correct answer is. In one example, the passport stamp costs €43 in French Guiana and €86 in mainland France.
The post walks through the workflow with a French passport-agency agent. The personas are used to generate synthetic conversations, and the facts in each answer are drawn from official procedure text. Training and evaluation run on NVIDIA's NeMo stack, with steps provided in the Nemotron repo.
In NVIDIA's example, a base model tells a user who can't travel for health reasons to apply by post, which the government's documentation says is not allowed. After fine-tuning, the model correctly describes the home-visit process. NVIDIA shows only this one before-and-after example and reports no benchmark results.
Pleias uses three Nemotron-Personas datasets, including the French and Belgian ones, for Lamina, a spine-care assistant built by one of its partners. "Better grounded data is what makes a small efficient regional-specific model viable," the company said. NVIDIA notes that persona grounding improves coverage of the training data but does not guarantee fairness or proper representation.