lab-ii/Aina-14B

Aina-14B-Instruct: A Yakut Language LLM

Aina-14B-Instruct is a 14 billion parameter Large Language Model developed by lab-ii, built upon the Qwen/Qwen3-14B architecture. This model has undergone significant continued pre-training and supervised fine-tuning (SFT) to specialize in the Yakut (Sakha) language, addressing the challenges of low-resource languages.

Key Capabilities

• Yakut Language Specialization: The model was pre-trained on the ailabykt/sakha-corpus-mono dataset and fine-tuned using lab-ii/sakha_chat_ml-sft, making it highly proficient in Yakut.
• Instruction Following: As an instruction-tuned model, Aina-14B-Instruct is designed to follow user prompts effectively in its specialized language.
• Resource Optimization: Training was conducted with a context length of 8000 tokens due to resource constraints, indicating an efficient use of available computational power for its specific goal.

Good For

• Yakut Language Applications: Ideal for use cases requiring natural language understanding and generation in Yakut, such as translation, content creation, or conversational AI.
• Low-Resource Language Research: Provides a valuable foundation for further research and development in AI for underrepresented languages.
• Telegram Bot Integration: An existing Telegram bot demonstrates its practical application in real-world scenarios, though without chat history due to context limitations.