The new model has been made available in three variants: a base version, which is used to fine-tune your own versions of the model, an instructional version, which can perform a variety of tasks, and a trained version, which is secured and adapted to conduct conversations with users.
The operational manager of the HIVE AI project, dr hab. Piotr Pęzik, Associate Professor at the University of Lodz, notes that:
From the beginning, we have maintained that the mass copying of pre-made AI models, the so-called strong LLMs, is associated with a number of risks. Therefore, we are building our own training data resources and developing a methodology for the controlled generation of synthetic data – data derived from other models but validated and validated by humans. Its initial results were used to fine-tune new 12B models.
Thanks to this, the model better understands the Polish cultural context and generates fewer redundant words. Therefore, its answers are more precise.
The new PLLuM model, as well as the entire HIVE AI project, will not only strengthen Polish digital sovereignty, but is also an investment in the future that will support the development of science and innovation in Poland, while building a more secure and efficient administration.
We have written about the earlier stages of work on PLLuM here:
- PLLuM, or AI in Polish: Have Poles Created Something Better than ChatGPT?
- Polish Language Model Is Ready! UniLodz Among the Creators of PLLuM
- Yet Another Success of the UniLodz Scientists – PLLuM Goes Global!
Source: NASK
Edit: Anna Stanecka, Press Office, University of Lodz
