GAI-LLM/ko-en-llama2-13b-mixed-v3

Overview

GAI-LLM/ko-en-llama2-13b-mixed-v3 is a 13 billion parameter auto-regressive language model built upon the LLaMA2 transformer architecture. Developed by Donghoon Oh, Hanmin Myung, and Eunyoung Kim from SK C&C G.AI Eng, this model specializes in processing and generating text in both Korean and English.

Key Capabilities

• Bilingual Proficiency: The model is specifically trained on a mixed dataset combining Open Korean Datasets (Kopen-platypus + kaist_cot_deepL) to enhance its performance in Korean and English language tasks.
• LLaMA2 Architecture: Leverages the robust LLaMA2 base model, specifically hyunseoki/ko-en-llama2-13b, providing a strong foundation for language understanding and generation.
• Text-to-Text Generation: Designed for general text input and text output applications.

Training Details

The model was trained using A100 GPU 80GB * 8, indicating a significant computational investment to achieve its bilingual capabilities.

Benchmarking

Performance can be tracked and compared on the Open KO-LLM LeaderBoard, providing transparency and a reference for its standing among other Korean LLMs.

Recommended Use Cases

This model is particularly well-suited for applications requiring strong performance in both Korean and English, such as bilingual chatbots, content generation, translation assistance, or any task where understanding and generating text in these two languages is crucial.