kyujinpy/Kosy-platypus2-13B-v2
kyujinpy/Kosy-platypus2-13B-v2 is a 13 billion parameter causal language model developed by Kyujin Han (kyujinpy), fine-tuned using the NEFTune method. Based on the ko-en-llama2-13b architecture, this model is optimized for Korean language understanding and generation, demonstrating competitive performance on various Korean language benchmarks. It is particularly suited for tasks requiring robust Korean natural language processing capabilities.
Loading preview...
Model Overview
kyujinpy/Kosy-platypus2-13B-v2, also known as Kosy🍵llama, is a 13 billion parameter language model developed by Kyujin Han (kyujinpy). It is built upon the hyunseoki/ko-en-llama2-13b base model and has been fine-tuned using the NEFTune method, which incorporates noisy embedding fine-tuning. The training utilized the kyujinpy/KOpen-platypus combined dataset, leveraging A100 GPU 40GB and COLAB resources.
Key Capabilities & Performance
This model is specifically designed for Korean language tasks. Benchmarks on the KO-LLM leaderboard show its performance across various Korean-specific evaluations:
- Ko-ARC
- Ko-HellaSwag
- Ko-MMLU
- Ko-TruthfulQA
- Ko-CommonGen V2
TheNEFT(🍵kosy)+MLP-v2variant achieved an average score of 45.45, with notable scores in Ko-HellaSwag (54.56) and Ko-CommonGen V2 (42.98). TheNEFT(🍵kosy)+MLP-v3variant, available askyujinpy/Kosy-platypus2-13B-v3, shows further improvements with an average of 46.31.
When to Use This Model
This model is ideal for applications requiring strong Korean language understanding and generation, especially where the benefits of NEFTune's fine-tuning approach are desired. Its performance on Korean benchmarks suggests suitability for tasks such as:
- Korean text generation
- Korean question answering
- Korean language understanding and reasoning tasks
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.