PLLuM

PLLuM (Polish Large Language Model) is a family of open large language models for Polish, developed in 2024 by a consortium of Polish research institutions led by Wrocław University of Science and Technology. The project was funded by the Polish Ministry of Digital Affairs (targeted grant no. 1/WI/DBiI/2023, PLN 14,504,392) with the goal of producing a model that supports the Polish public and private sectors, with explicit focus on the Polish language and cultural context.

The family includes base, instruct, and chat variants at 8B, 12B, 8x7B (MoE), and 70B parameter scales. Variants tagged "nc" are released under CC BY-NC 4.0 (non-commercial); others inherit the licenses of their base models (Llama 3.1 Community License for Llama-PLLuM variants, Apache 2.0 for some others). Models are produced by continued pretraining + SFT + preference tuning on top of existing architectures — Llama 3.1 (8B, 70B), Mistral 7B, and Mixtral 8x7B — using a large Polish text corpus.

The project formally concluded on 31 December 2024; follow-up work continues under the HIVE AI initiative (NASK). Models are publicly available on HuggingFace under the CYFRAGOVPL organization (Polish Ministry of Digital Affairs), and an interactive chat is hosted at pllum.clarin-pl.eu by Wrocław University of Science and Technology.