UNIVA-Bllossom/DeepSeek-llama3.1-Bllossom-8B

The DeepSeek-llama3.1-Bllossom-8B model, developed by UNIVA and Bllossom, is an 8 billion parameter language model built upon the DeepSeek-R1-Distill-Llama-8B base. It is specifically optimized to enhance reasoning performance in Korean environments, addressing limitations of its base model which was primarily trained on English and Chinese data. This model achieves improved Korean inference by performing internal reasoning in English and then generating responses in the user's input language.