Llama-3.2-3B-F1-Instruct (Formosa-1)
| Property | Value |
|---|---|
| Model Type | LlamaForCausalLM |
| Languages | Traditional Chinese & English |
| License | llama3.2 |
| Authors | Huang Liang Hsun, Min Yi Chen, Wen Bin Lin, Chao Chun Chuang & Dave Sung |
What is Llama-3.2-3B-F1-Instruct?
Llama-3.2-3B-F1-Instruct, also known as Formosa-1 or F1, is a specialized language model developed through collaboration between Twinkle AI and APMIC, with technical guidance from the National Center for High-Performance Computing in Taiwan. This model represents a significant advancement in Traditional Chinese language processing, specifically tailored for Taiwan-specific contexts and requirements.
Implementation Details
The model builds upon the LlaMA architecture and has been fine-tuned to excel in various domain-specific tasks. Evaluation results show significant improvements over the base Llama-3.2-3B-Instruct model, particularly in TMMLLU+ (42.18%) and Taiwan legal domain tasks (31.26%). The model also demonstrates strong performance in mathematical reasoning (MATH-500: 51.40%) and general question-answering tasks (GPQA Diamond: 33.84%).
- Enhanced instruction-following capabilities
- Optimized for Traditional Chinese language understanding
- Specialized in legal, educational, and daily-life applications
- Improved performance on Taiwan-specific tasks
Core Capabilities
- Strong performance in Traditional Chinese language processing
- Enhanced mathematical reasoning abilities
- Specialized legal domain knowledge for Taiwan context
- Robust instruction-following capabilities
Frequently Asked Questions
Q: What makes this model unique?
The model's specialization in Traditional Chinese and Taiwan-specific contexts, combined with its strong performance in legal and educational domains, makes it particularly valuable for applications in Taiwan. Its instruction-tuning optimization sets it apart from general-purpose language models.
Q: What are the recommended use cases?
The model is well-suited for legal document processing, educational applications, and general-purpose Traditional Chinese language tasks. It shows particular strength in mathematical reasoning and can be effectively used in both academic and professional contexts within Taiwan.
