Model Overview

sh2orc/Llama-3.1-Korean-8B-Instruct is an 8 billion parameter instruction-tuned model, built upon the Meta-Llama-3.1-8B-Instruct architecture. It has been specifically fine-tuned to enhance its performance in the Korean language, making it a specialized tool for Korean-centric AI applications. The model supports a substantial context length of 32768 tokens, allowing for processing and generating longer, more coherent Korean texts.

Key Capabilities

• Korean Language Proficiency: Optimized for understanding and generating high-quality Korean text.
• Instruction Following: Capable of adhering to complex instructions for various tasks.
• Question Answering: Demonstrates strong performance in answering questions based on provided context, as shown in examples covering historical figures, global events, and practical advice.
• Multilingual Translation: Can perform translation tasks, specifically demonstrated with English-Korean and Korean-English translations.
• Conversational AI: Designed for interactive dialogue, providing informative and contextually relevant responses.

Training Data

The model's Korean language capabilities are significantly bolstered by its training on a diverse set of Korean datasets, including:

• maywell/ko_wikidata_QA
• lcw99/wikipedia-korean-20240501-1million-qna
• jojo0217/korean_rlhf_dataset
• MarkrAI/KoCommercial-Dataset

Usage

This model is readily usable with popular libraries like transformers and vllm, providing flexible deployment options for developers. Example code snippets are provided for both frameworks, illustrating how to set up text generation pipelines and apply chat templates for conversational inference.