We actively participate in the development of two Polish Large Language Models (LLMs): Bielik and PLLuM (HIVE AI). Polish language models are trained on Polish data and therefore are better adapted to the Polish language than foreign models. This significantly facilitates their use in various sectors, such as education, business, or public administration. From a broader perspective, the development of Polish language models can help protect and promote the Polish language, which is important for national and cultural identity.
LLM development requires enormous computational resources, mainly due to the complexity of tasks that must be performed during training. The key infrastructure is one that provides many modern graphics processors (GPUs), as they allow for faster processing of text sets numbering in billions of words, as well as performing matrix operations of sizes in the order of millions or billions of elements, compared to CPUs. Additionally, machine learning libraries such as TensorFlow and PyTorch are optimized for GPU usage. This allows for the full potential of these processors to be utilized in training language models.
For this reason, Cyfronet provides resources of its supercomputers - in particular, Helios - for training Bielik and PLLuM. Additionally, Cyfronet experts support the process substantively.
Bielik
Bielik is a family of Polish language models that was created through the collaboration of the SpeakLeash Foundation and ACK Cyfronet AGH. Bielik comes in various versions, ranging from 1.5 to 11 billion parameters.
SpeakLeash – group of enthusiasts and creators of Bielik
The SpeakLeash Foundation brings together people from various professions who aim to create the largest Polish text dataset. The project team consists of employees from Polish companies, researchers from scientific centers, and students of AI-related fields. Work on the first officially and widely released model (Bielik-11B-v2) took over a year and involved collecting, processing, and classifying data. In 2025, versions 2.5 and 3 were made available.
Open model and availability for business use
Bielik is an open-source project and can therefore be used for both private and business purposes. It can be used to support business development and business processes. Examples of applications include:
- Customer service automation: chatbots, answering frequently asked questions.
- Data analysis: processing and analysis of large datasets, generating reports.
- Content creation: articles, product descriptions, marketing materials.
- Support in decision-making processes: scenario generation, risk analysis, recommendations.
Bielik available in PLGrid
Scientists affiliated with Polish research institutions can use Bielik as part of the services available through the PLGrid Portal.
You are invited to visit the following sites:
PLLuM (HIVE AI)
In 2025, Cyfronet joined the HIVE consortium, which was established as a continuation of the PLLuM consortium, developing a family of language models with the same name. PLLuM is being developed at the request of the Ministry of Digitization. Versions using 8 to 70 billion parameters have been released, providing a high degree of flexibility in application.
The HIVE Consortium
8 leading units ara currently engaged in the HIVE AI project development:
- Wrocław University of Science and Technology (leader)
- Institute of Computer Science, Polish Academy of Sciences
- Institute of Slavic Studies, Polish Academy of Sciences
- Research and Academic Computer Network (NASK)
- National Information Processing Institute
- University of Łódź
- Centre for Information Technology
- ACC Cyfronet AGH
Ethical and for society
Like Bielik, PLLuM is based on ethically sourced data – the training data was used in accordance with copyright regulations. Additionally, data directly obtained from owners who granted a license for their use was used in the project. The PLLuM model is, by its nature, a safe and harm-free model. It is planned to be widely made available to society in the form of an Assistant integrated into the mObywatel application. In this way, PLLuM is intended to facilitate access to regulations and provide more convenient handling of official matters.
Meet the PLLuMs on Hugging Face