Overview

komt-Llama-2-13b-hf is a 13 billion parameter auto-regressive language model developed by davidkim (changyeon kim). It is built upon the Llama 2 architecture and has been specifically fine-tuned using a multi-task instruction technique with a Korean multi-task instruction dataset. This specialization aims to significantly improve its performance and utility for Korean language processing tasks.

Key Capabilities

• Enhanced Korean Language Performance: Achieves improved accuracy on Korean Semantic Textual Similarity benchmarks, outperforming base Llama-2-7b-chat-hf and other Korean Llama-2 variants.
• Optimized Transformer Architecture: Utilizes an optimized transformer architecture, fine-tuned with supervised fine-tuning (SFT) via multi-task instruction.
• Instruction Following: Designed to follow instructions effectively, as demonstrated by its prompt template and example responses in Korean.

Good For

• Applications requiring robust Korean language generation and understanding.
• Developing chatbots or conversational AI systems in Korean.
• Research and development focused on improving LLM performance for specific non-English languages, particularly Korean.