Teuta: Bilingual Instruction-Tuned Language Model

Teuta, developed by LTS-VVE, is a 3.2 billion parameter instruction-tuned language model built upon the meta-llama/Llama-3.2-3B base. Its primary focus is bilingual question answering and instruction-following in both Albanian (sq) and English (en), making it particularly valuable for multilingual applications and supporting under-resourced languages.

Key Capabilities & Features

• Bilingual Proficiency: Optimized for instruction-following and question answering in Albanian and English.
• Diverse Domain Knowledge: Fine-tuned on a broad spectrum of subjects including mathematics, philosophy, chemistry, biology, code (with a specific emphasis on Rust), psychology, and climate science.
• Instruction Following: Designed to handle various instructional prompts, from academic queries to more open-ended tasks.
• Generalization: Leverages both synthetic and real datasets to enhance its ability to generalize across technical and non-technical domains.

Ideal Use Cases

• Research: Suitable for exploring language model capabilities in bilingual contexts.
• Educational Tools: Can be integrated into tools for learning or information retrieval in Albanian and English.
• Domain-Specific Applications: Effective for applications requiring specialized knowledge in the aforementioned scientific and technical fields.
• Under-resourced Language Support: Particularly strong for applications focusing on the Albanian language.

Important Considerations

• The model's training data includes sensitive content (e.g., mental health, therapy, philosophical questions).
• Outputs are not guaranteed to be factual or safe, requiring careful consideration for sensitive applications.