domestic-yak-8B-instruct: A Macedonian Instruction-Tuned LLM

LVSTCK/domestic-yak-8B-instruct is an 8 billion parameter language model developed by LVSTCK, specifically fine-tuned for instruction-following in the Macedonian language. This model builds upon the domestic-yak-8B base and was trained for three epochs on the sft-mk dataset, which comprises approximately 100k samples covering various categories like QA, chat, reasoning, essays, and code.

Key Capabilities

• Macedonian Language Proficiency: Optimized for generating coherent and task-specific responses in Macedonian.
• Instruction Following: Enhanced capabilities for understanding and executing user instructions, making it suitable for interactive applications.
• Competitive Performance: Benchmarks show it performs on par with, and in some cases surpasses, larger models like Llama 70B on Macedonian-specific evaluations, establishing it as a leading 8B parameter model for the language.

Good for

• Developing chatbots and virtual assistants for Macedonian speakers.
• Applications requiring task-specific responses and instruction adherence in Macedonian.
• Research and development in low-resource language NLP, particularly for Macedonian.