KAERI-MLP/llama-3.1-Korean-AtomicGPT-Bllossom-8B

KAERI-MLP/llama-3.1-Korean-AtomicGPT-Bllossom-8B Overview

This model, developed by KAERI-MLP, is an 8 billion parameter language model built upon the Llama 3.1 architecture. It is specifically designed and fine-tuned for Korean language processing, with a specialized focus on atomic energy (원자력) related content. The model supports a substantial context length of 32,768 tokens, enabling it to handle complex and detailed queries.

Key Capabilities

• Specialized Korean Language Processing: Optimized for understanding and generating Korean text.
• Atomic Energy Domain Expertise: Demonstrates strong performance in answering questions related to nuclear science and technology.
• Instruction Following: Capable of acting as a helpful AI assistant, providing kind and informative responses.
• Llama 3.1 Foundation: Benefits from the robust architecture of the Llama 3.1 series.

Performance

Evaluation results highlight the model's proficiency in the atomic energy domain:

• Atom 8B: Achieved an overall score of 8.09, with single-turn performance at 8.12 and multi-turn at 8.06 in atomic energy specific evaluations.
• Bllossom 8B: Showed even stronger performance with an overall score of 8.81, scoring 8.75 for single-turn and 8.88 for multi-turn interactions in the atomic energy category.

Good For

• Applications requiring detailed and accurate information on atomic energy in Korean.
• Chatbots or virtual assistants designed for Korean-speaking users with technical inquiries.
• Research and educational tools focused on nuclear science in the Korean language.