Dorna-Llama3-8B-Instruct Overview

Dorna-Llama3-8B-Instruct is an 8 billion parameter instruction-tuned model developed by Part AI, specifically fine-tuned on Persian data. It is built upon the robust Meta Llama 3 Instruct architecture, enhancing its capabilities for Persian language tasks.

Key Capabilities & Performance

This model has undergone extensive evaluation, both human and automatic (using GPT-4 as a judge), across a diverse set of tasks including:

• Boolean Questions (Complex and Easy)
• Code Generation
• Long Response (General and Historical)
• Math (Complex and Easy)
• News QA (Complex and Easy)
• Paraphrasing
• General Knowledge (Easy and Hard)
• Summarization

Notably, in human evaluations, Dorna-8B-it demonstrates strong performance, winning against "Persian Mind 7B" in 55.77% of cases and showing competitive results against "Meta-Llama-3-8B-Instruct" and "GPT 3.5 turbo-1106".

Automatic evaluations further highlight its strengths, with Dorna-8B-it achieving high win rates against several other models, including "Llama 3 base" (58.96% overall win rate), "Part Mistral 7B" (77.20%), "Persian Mind 7B" (90.88%), and "PersianLlama 7B" (98.70%).

Ideal Use Cases

This model is particularly well-suited for applications requiring high-quality Persian language understanding and generation, especially in conversational AI, content creation, and information retrieval within a Persian context. Its fine-tuning on Persian data makes it a strong candidate for tasks where other general-purpose models might underperform in this specific language.