CYFRAGOVPL/Llama-PLLuM-70B-instruct-2412

PLLuM: Polish Large Language Models

CYFRAGOVPL/Llama-PLLuM-70B-instruct-2412 is a 70 billion parameter instruction-tuned model from the PLLuM family, developed by CYFRAGOVPL. It is built upon Llama 3.1 and specialized for Polish, Slavic, and Baltic languages, incorporating English data for enhanced generalization. The model was refined using an extensive, manually curated Polish "organic instructions" dataset (~40k prompt-response pairs, including ~3.5k multi-turn dialogs) and the first Polish-language preference corpus for alignment.

Key Capabilities

• Multilingual Proficiency: Strong performance in Polish, Slavic, and Baltic languages, with English generalization.
• Instruction Following: Fine-tuned with a unique, human-authored instruction set to cover subtle aspects of supervised fine-tuning.
• Preference Learning: Aligned using a Polish preference corpus to ensure correctness, balance, and safety, especially for sensitive topics.
• Domain-Specific Excellence: Achieves top scores on custom benchmarks for Polish public administration tasks.
• Retrieval Augmented Generation (RAG): Trained to perform well in RAG settings, providing answers based on provided documents with citations.

Good For

• General Language Tasks: Text generation, summarization, and question answering in Polish.
• Domain-Specific Applications: Developing intelligent assistants for Polish public administration, legal, or bureaucratic topics.
• Research & Development: As a foundational model for AI applications requiring strong Polish language capabilities.