CYFRAGOVPL/Llama-PLLuM-70B-chat-2508
CYFRAGOVPL/Llama-PLLuM-70B-chat-2508 is a 70 billion parameter large language model developed by the HIVE AI Consortium, specialized in Polish and other Slavic/Baltic languages, with a context length of 32768 tokens. Built on the Llama 3.1 architecture, it is fine-tuned with extensive Polish instruction and preference datasets to generate contextually coherent text and assist in tasks like question answering and summarization. This model excels in Polish-language tasks and is particularly effective for applications in Polish public administration, offering state-of-the-art results in its domain. It is aligned to human preferences for safer and more efficient dialogue in general-purpose scenarios.
Loading preview...
Overview
CYFRAGOVPL/Llama-PLLuM-70B-chat-2508 is a 70 billion parameter large language model developed by the HIVE AI Consortium, specializing in Polish and other Slavic/Baltic languages. It incorporates additional English data for broader generalization and is built upon the Llama 3.1 architecture. The model is extensively refined through instruction tuning, preference learning, and advanced alignment techniques, making it suitable for generating contextually coherent text and assisting in various tasks like question answering and summarization.
Key Capabilities
- Specialized Multilingualism: Optimized for Polish, Slavic, and Baltic languages, with a strong command of Polish. It was pretrained on up to 150 billion tokens of Polish corpora.
- High-Quality Instruction Tuning: Fine-tuned with approximately 55,000 manually curated Polish "organic instructions" and additional programmatic and synthetic instructions, covering a wide range of human-model interactions.
- Robust Alignment: Features the first Polish-language preference corpus, manually assessed for correctness, balance, and safety across seven criteria (truthfulness, linguistic correctness, safety, fairness, conciseness, coherence & reasoning, helpfulness & instruction-following).
- Domain-Specific Excellence: Achieves top scores on custom benchmarks relevant to Polish public administration and state-of-the-art results in broader Polish-language tasks.
- Dialogue Optimization: The
-chatvariant is specifically aligned to human preferences, making it safer and more efficient for general-purpose dialogue and conversational AI.
Good For
- General Language Tasks: Ideal for text generation, summarization, and question answering in Polish.
- Domain-Specific Assistants: Particularly effective for applications within Polish public administration, legal, or bureaucratic contexts, especially when paired with Retrieval Augmented Generation (RAG).
- Research & Development: Serves as a strong foundation for building downstream AI applications that require robust Polish language capabilities.