kyujinpy/Kosy-platypus2-13B-v5

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

kyujinpy/Kosy-platypus2-13B-v5 is a 13 billion parameter Llama 2-based model developed by Kyujin Han, fine-tuned using the NEFTune method for improved performance. This model is specifically optimized for Korean language tasks, leveraging a ko-en-llama2-13b base and the KOpen-platypus dataset. It demonstrates competitive performance on Korean LLM benchmarks, particularly in areas like Ko-CommonGen V2.

Loading preview...

Model Overview

kyujinpy/Kosy-platypus2-13B-v5, also known as Kosy🍵llama, is a 13 billion parameter language model developed by Kyujin Han. It is built upon the hyunseoki/ko-en-llama2-13b base model and fine-tuned using the NEFTune method, which incorporates noisy embedding fine-tuning. The training utilized the kyujinpy/KOpen-platypus combined dataset.

Key Capabilities

  • Korean Language Proficiency: Optimized for Korean language understanding and generation, as evidenced by its performance on various Ko-LLM benchmarks.
  • NEFTune Integration: Leverages the NEFTune method, which can be explored further via the associated KoNEFTune GitHub repository.
  • Benchmark Performance: Achieves an average score of 46.31 on the KO-LLM leaderboard, with notable scores in Ko-CommonGen V2 (46.16) and Ko-HellaSwag (54.54) for its v3 iteration.

Good For

  • Korean NLP Applications: Ideal for tasks requiring strong performance in the Korean language.
  • Research on NEFTune: Provides a practical example of NEFTune application in a large language model context.
  • Comparative Analysis: Useful for comparing performance against other Korean LLMs, particularly those listed on the KO-LLM leaderboard.

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