Renaissance Labs

Atlas / Europe

Community collective · Germany

Occiglot

An informal European research collective that released Occiglot-7B models, multilingual web datasets and tokenizer analyses for European languages.

language modelsEuropean languagesdatasetsevaluation

Germany

Founded
2024
Type
Community collective
Based in
Germany
Key papers
1
Featured projects
8

About

Occiglot describes itself as an ongoing research collective and an informal community of motivated individuals building open multilingual language models for Europe. Its announcement in March 2024 framed the work as preserving linguistic diversity and AI sovereignty in Europe.

The group released continually pre-trained Occiglot-7B models for German, French, Italian, Spanish and combined European language sets in March 2024, followed by the Occiglot-Fineweb dataset and German models built with DiscoResearch. It also published a tokenizer evaluation for European languages and contributed to the Community OSCAR web corpus.

The page names contributors Manuel Brack, Malte Ostendorff and Pedro Ortiz Suarez, and invites collaborators through its Discord server.

Mission and ambitions

Occiglot works to build transparent, open-source multilingual language models for European languages through an open community.

Known for

  1. 01

    Occiglot-7B multilingual European models

  2. 02

    Occiglot-Fineweb dataset

  3. 03

    European language tokenizer evaluation

  4. 04

    Community OSCAR web data

Milestones

  1. 2024

    Occiglot announcement

    The collective published its mission post on 6 March 2024.

  2. 2024

    Occiglot-7B model family

    Language-specific and eu5 Occiglot-7B models appeared on Hugging Face in early March 2024.

  3. 2024

    Tokenizer evaluation

    An analysis of tokenizers on European languages published on 11 March 2024.

  4. 2024

    Occiglot-Fineweb

    A preliminary dataset of about 230 million cleaned documents in 10 languages.

  5. 2024

    German Llama-3 models

    Base, instruct and long-context German models built with DiscoResearch on 65 billion German tokens.

  6. 2024

    Community OSCAR

    A multilingual web dataset built from 40 monthly Common Crawl snapshots, announced in August 2024.

Open source

GitHubgithub.com/occiglot
Hugging Facehuggingface.co/occiglot

Research papers

Notable releases

Openness

Models, datasets and evaluation code are public on Hugging Face and GitHub.

People

Listed from the lab's own pages and public profiles.

Contact and community

Occiglot's site invites contact and Discord joining from people working on multilingual LLMs.

Discord of the informal collective for discussing multilingual LLM work; the site invites newcomers and collaborators to join.

Sources

Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.

Labs nearby