README.md

metadata

title: README
emoji: 👀
colorFrom: gray
colorTo: purple
sdk: static
pinned: false

Diabolocom Research

The research lab of Diabolocom, building speech and language models for real-world conversations.

We release our models, datasets, and code openly.

Selected work

• BidirLM — multilingual bidirectional encoders across text, audio, and image, with an omnimodal 2.5B variant supporting 119 languages (Apache 2.0)
• EuroBERT — multilingual encoder suite (210M / 610M / 2.1B), developed with CentraleSupélec (MICS), Artefact, and Unbabel (Apache 2.0)

Website · Research · GitHub