Deniss8686/YugoGPT

YugoGPT: A Specialized LLM for Bosnian, Croatian, and Serbian

YugoGPT is a 7 billion parameter base Large Language Model (LLM) developed by Aleksa Gordić, specifically designed and optimized for Bosnian, Croatian, and Serbian (BCS) languages. Built on the robust Mistral 7B architecture, this model has been extensively trained on tens of billions of BCS tokens, positioning it as a leading open-source solution for these languages.

Key Capabilities and Features

• Multilingual Specialization: Primarily focused on generating high-quality text in Bosnian, Croatian, and Serbian.
• Base Model Architecture: Functions as a powerful autocomplete engine, providing text completions rather than following complex instructions or having built-in moderation.
• Performance: Evaluation results against models like Mistral 7B, LLaMA 2 7B, and GPT2-orao on Serbian language tasks demonstrate its strong performance in the BCS linguistic domain. These evaluations were conducted using the serbian-llm-eval framework.

Intended Use Cases

• Research and Development: Ideal for researchers and developers working on natural language processing tasks specific to BCS languages.
• Text Generation: Suitable for applications requiring text completion or generation in Bosnian, Croatian, or Serbian.
• Foundation for Fine-tuning: As a base model, YugoGPT can serve as an excellent foundation for further fine-tuning to create instruction-following or moderated models for specific BCS applications. More powerful, instruction-tuned iterations are available via RunaAI's API platform.

Limitations

As a base model, YugoGPT does not include moderation mechanisms and is not designed to follow instructions directly. Users seeking instruction-tuned or moderated BCS LLMs should consider the more advanced models available through RunaAI's API.