CYFRAGOVPL/PLLuM-4B-chat-2512
The CYFRAGOVPL/PLLuM-4B-chat-2512 is a 4.3 billion parameter instruction-tuned causal language model developed by the PLLuM consortium and HIVE AI, based on Google's Gemma-3-4b-pt architecture. This model is specialized in Polish language tasks, incorporating English data for generalization, and is refined through extensive instruction tuning and preference learning. It excels at generating contextually coherent text and assisting in various tasks, particularly within Polish public administration contexts.
Loading preview...
PLLuM-4B-chat-2512: Polish Language Model
CYFRAGOVPL/PLLuM-4B-chat-2512 is a 4.3 billion parameter model from the PLLuM family, developed by the PLLuM consortium and HIVE AI. It is built upon Google's Gemma-3-4b-pt and is specifically aligned for human preferences, making it suitable for dialogue and general-purpose scenarios. The model's development involved extensive data collection, including a large-scale, high-quality Polish and English text corpus, rigorously cleaned and deduplicated.
Key Capabilities
- Specialized Polish Language Processing: Optimized for Polish, with additional English data for broader generalization.
- Organic Instruction Tuning: Fine-tuned on approximately 70k manually curated Polish "organic instructions" and an additional 90k programmatically derived and synthetic instructions, mitigating negative linguistic transfer.
- Preference Learning: Utilizes a ~60k manually annotated Polish preference corpus to ensure balanced, safe, and contextually appropriate responses, even for sensitive topics.
- Strong Performance: Achieves state-of-the-art results in broader Polish-language tasks and top scores on custom benchmarks relevant to Polish public administration.
- RAG Optimization: Designed to perform well in Retrieval-Augmented Generation (RAG) settings, providing document-cited answers or indicating when information is unavailable.
Good For
- General Language Tasks: Text generation, summarization, extraction, and question answering in Polish.
- Domain-Specific Assistants: Particularly effective for applications within Polish public administration, legal, and bureaucratic domains.
- Research & Development: Serving as a foundational model for downstream AI applications requiring strong Polish language command.