TurkuNLPBLUUMI

BLUUMI 176B

BLUUMI adds Finnish capability to BigScience's multilingual BLOOM 176B model through continued pretraining. It is a trained BLOOM derivative, separate from the Finnish-only FinGPT series.

Model design and training

The model continues training for 40 billion tokens on ROOTS and Finnish data. The Finnish material includes web collections, National Library resources, Yle and STT news, and discussion archives. ROOTS retains material from BLOOM's original multilingual training. The release provides a base model rather than a documented instruction-tuned assistant.bloom-finnish-176b: official model card

Practical assessment

Practical guidance. BLUUMI offers a useful comparison between adapting a large multilingual model and training a smaller monolingual model. Plan deployment around its substantial weight memory. Evaluate Finnish gains alongside other languages if retaining multilingual behavior matters to the project.

Specifications & access

BLUUMI 176B specifications
DetailValue and conditions
AccessDownloadable
Model identifiersbloom-finnish-176bBLOOM + Finnish 176B
Model size176 billion parameters.bloom-finnish-176b: official model card
LicenseBigScience BLOOM RAIL 1.0.bloom-finnish-176b: official model card
Parent modelBigScience BLOOM.bloom-finnish-176b: official model card

Related models

  • FinGPT 13B

    TurkuNLP ยท FinGPT

    FinGPT 13B is a Finnish text-generation base model in TurkuNLP's Finnish GPT-3 series.

    Launch date not confirmedDownloadable