kyujinpy/Korean-OpenOrca-13B

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kLicense:cc-by-nc-sa-4.0Architecture:Transformer0.0K Open Weights Warm

Korean-OpenOrca-13B is a 13 billion parameter auto-regressive language model developed by Kyujin Han, based on the LLaMA2 transformer architecture. It is fine-tuned using the OpenOrca-KO dataset, a Korean translation of OpenOrca, to enhance its performance on Korean language tasks. This model excels in various Korean benchmarks, offering strong capabilities for general-purpose Korean text generation and understanding.

Loading preview...

Overview

Korean-OpenOrca-13B is a 13 billion parameter language model developed by Kyujin Han (kyujinpy), built upon the LLaMA2 transformer architecture. It was created through a research consortium involving (주)미디어그룹사람과숲 and (주)마커. The model is specifically fine-tuned for Korean language processing, utilizing the OpenOrca-KO dataset, which is a DeepL translation of the original OpenOrca dataset.

Key Capabilities

  • Korean Language Proficiency: Optimized for understanding and generating Korean text.
  • LLaMA2 Architecture: Benefits from the robust and widely-used LLaMA2 base.
  • Benchmark Performance: Achieves a competitive average score of 47.85 on the KO-LLM leaderboard, with notable scores in Ko-HellaSwag (54.13) and Ko-CommonGen V2 (56.57).
  • Comparative Strength: Outperforms its base model, hyunseoki/ko-en-llama2-13b, across several Korean benchmarks.

Use Cases

This model is suitable for a variety of Korean natural language processing tasks, including:

  • General text generation in Korean.
  • Question answering and conversational AI in Korean.
  • Content creation and summarization for Korean text.

Licensing

The model is released under the cc-by-nc-sa-4.0 license.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p