kyujinpy/KOR-Orca-Platypus-13B-v2
KOR-Orca-Platypus-13B-v2 is a 13 billion parameter auto-regressive language model developed by Kyujin Han (kyujinpy) in a research consortium with Media Group Saramgwasup and Marker Inc. Based on the LLaMA2 transformer architecture, it is fine-tuned for Korean language tasks, leveraging the hyunseoki/ko-en-llama2-13b base model. This model demonstrates improved performance on Korean language benchmarks, making it suitable for applications requiring strong Korean language understanding and generation.
Loading preview...
KOR-Orca-Platypus-13B-v2 Overview
KOR-Orca-Platypus-13B-v2 is a 13 billion parameter auto-regressive language model developed by Kyujin Han (kyujinpy) as part of a research consortium. It is built upon the LLaMA2 transformer architecture, specifically utilizing the hyunseoki/ko-en-llama2-13b as its base model.
Key Capabilities & Training
This model is designed for Korean language processing, having been fine-tuned on a private dataset, kyujinpy/KOR-OpenOrca-Platypus-v3. The training was conducted using A100 GPU 40GB and COLAB resources. It processes text inputs and generates text outputs.
Performance Benchmarks
KOR-Orca-Platypus-13B-v2 shows competitive performance on the KO-LLM leaderboard, which tracks various Korean language benchmarks. Compared to its predecessor, KOR-Orca-Platypus-13B, the v2 model demonstrates an improved average score of 49.48, up from 46.59. Notable scores include:
- Ko-ARC: 44.03
- Ko-HellaSwag: 54.43
- Ko-CommonGen V2: 65.05
These results indicate its proficiency across different Korean language understanding and generation tasks, positioning it among the top models on the Open KO-LLM LeaderBoard as of its last update.
Licensing
The model is released under the cc-by-nc-sa-4.0 license.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.