jingyeom/SOLAR_KO_1.3_deup
The jingyeom/SOLAR_KO_1.3_deup model is a 15 billion parameter causal language model based on the beomi/OPEN-SOLAR-KO-10.7B architecture, fine-tuned using a deduplicated public dataset. It is specifically optimized for Korean language tasks, demonstrating strong performance across various Korean benchmarks. This model is suitable for applications requiring robust Korean language understanding and generation capabilities.
Loading preview...
Model Overview
jingyeom/SOLAR_KO_1.3_deup is a 15 billion parameter language model built upon the beomi/OPEN-SOLAR-KO-10.7B base model. It has been further trained using a collection of public datasets, with a focus on data quality through a deduplication algorithm, which is known to enhance language model performance.
Key Capabilities & Performance
This model is particularly strong in Korean language understanding and generation, as evidenced by its performance on the Ko-LLM-Leaderboard. As of January 29, 2024, it ranked 11th on the leaderboard, achieving an average score of 53.63. Its benchmark results include:
- Ko-ARC: 52.65
- Ko-HellaSwag: 60.92
- Ko-MMLU: 50.9
- Ko-TruthfulQA: 45.14
- Ko-CommonGen V2: 58.56
Use Cases
Given its strong performance on Korean benchmarks, this model is well-suited for applications requiring high-quality Korean text processing. This includes tasks such as:
- Korean language generation
- Question answering in Korean
- Text summarization for Korean content
- General Korean natural language understanding tasks