
Academic lab · Ljubljana, Slovenia
CJVTUniversity of Ljubljana
The University of Ljubljana centre that builds Slovene language resources, corpora and the open GaMS generative models.
Ljubljana, Slovenia
- Type
- Academic lab
- Based in
- Ljubljana, Slovenia
- Key papers
- 2
- Featured projects
- 8
About
CJVT, the Centre for Language Resources and Technologies, is a research unit of the University of Ljubljana with its organizational base at the Faculty of Computer and Information Science and five other participating faculties. It develops and maintains digital language resources for contemporary Slovene and makes tools available free of charge.
Its resources include the Sloleks morphological lexicon, which invites users to vote on accentuation and pronunciation, the ssj500k training corpus and its successor SUK, and the Tolmac system for translating lectures from Slovene. The centre joined the ELEXIS lexicography network in 2020 and works within a national program on language resources and technologies for Slovene that runs from 2019 to 2027.
The GaMS (Generative Model for Slovene) family began in 2024 with a 1 billion parameter model and has grown to 2B, 9B and 27B releases built on Gemma. A March 2026 preprint presents GaMS3-12B, adapted from Gemma 3 with continual pretraining on about 140 billion tokens and an open OCR pipeline for digitizing older Slovene texts.
Mission and ambitionsCJVT aims to build a long-term infrastructure of Slovene language resources and technology and to keep its tools and data openly available.
Known for
- 01
Slovene corpora and lexicons
- 02
GaMS Slovene language models
- 03
Sloleks and ssj500k
- 04
Open language technology for a smaller language
Milestones
- 2019
National language technology program
CJVT began work in the national program on language resources and technologies for Slovene, running from 2019 to 2027.
- 2020
ELEXIS membership
The centre joined the European lexicography infrastructure network.
- 2024
GaMS 1B
Published the first Generative Model for Slovene with 1 billion parameters (arXiv 2410.06898).
- 2026
GaMS3-12B
Presented a 12 billion parameter Slovene model adapted from Gemma 3, with an open OCR pipeline.
Open source
Fork of Stanford NLP Stanza for many languages, used in CLASSLA (53 stars)
Tagger and lemmatiser for Croatian, Serbian and Slovene (35 stars)
Geo-related BERT work (20 stars)
Slovene automatic speech recognition tool (20 stars)
12B Slovene instruct model, the most downloaded CJVT model.
9B Slovene instruct model.
- cjvt/slovenian-llm-evaldataset
Evaluation data for Slovenian LLMs.
- cjvt/sloleksdataset
Slovene lexicon data.
Research papers
Notable releases
Openness
Models and corpora are public on Hugging Face and GitHub, with licences set per project.
People
- Simon KrekCentre contact and corpus author
Listed as contact on the CJVT site and author of Slovene corpora.
- Marko Robnik-SikonjaProfessor, GaMS lead author
Co-author of the GaMS papers at the Faculty of Computer and Information Science.
- Domen VresResearcher, GaMS first author
First author of the GaMS 1B paper.
Listed from the lab's own pages and public profiles.
Contact and community
The CJVT website lists organisation, staff and news pages; its GitHub presence is the clarinsi organisation.
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.



