psyche/kollama2-7b: Korean-Optimized Llama-2 Model
psyche/kollama2-7b is a 7 billion parameter language model built upon the Llama-2 architecture, distinguished by its further training on extensive Korean datasets. This specialized training aims to enhance its proficiency and performance in processing and generating content in the Korean language.
Key Capabilities & Performance
The model's performance has been evaluated on the Open LLM Leaderboard, showcasing its capabilities across various benchmarks. Notable scores include:
- HellaSwag (10-shot): 78.78
- Winogrande (5-shot): 73.95
- ARC (25-shot): 53.24
- MMLU (5-shot): 42.31
These metrics indicate its general language understanding and reasoning abilities, with a specific focus on its performance within a Korean-centric context due to its training regimen.
Good For
- Korean Language Applications: Ideal for tasks requiring robust understanding and generation in Korean.
- Specialized NLP: Suitable for developers building applications where Korean language proficiency is a primary requirement, such as chatbots, content generation, or translation support for Korean speakers.
- Llama-2 Ecosystem Users: Offers a Korean-focused variant for those already familiar with or utilizing the Llama-2 model family.