AI Potluck
Model components / Benchmark / eval datasets

Compar:IA Datasets

Ministere de la Culture (France)

The Compar:IA datasets are human-preference and conversation data released from the French Government's Compar:IA LLM arena, published on Hugging Face by the Ministere de la Culture (SNUM). The collection includes comparia-votes (150,000+ conversation-level pairwise preference annotations), comparia-conversations, and comparia-reactions. The data is ~89% French and is intended for finetuning, alignment, and benchmarking of multilingual LLMs.

Verified live 2026-06-22 via primary sources. Open license and documented, but gated access requiring agreement caps it at gated.

Openness

3 high confidence
3.0
license
Etalab-2.0(permissive French-gov open license)
gated
HF contact-info agreement required

Open license and documented, but gated access requiring agreement caps it at gated.

Adoption

2 high confidence
2.0

comparia-votes ~155 and comparia-conversations ~202 downloads last month on Hugging Face.

Capability

not assessed

a dataset is not 'capable'

Unchanged since 2026-06-22 (last edited, not re-checked)