Model Overview

Saxo/Linkbricks-Horizon-AI-Korean-Gemma-2-sft-dpo-27B is a 27 billion parameter Korean language model developed by Yunsung Ji (Saxo), a data scientist at Linkbricks Horizon-AI. It is fine-tuned from the gemma-2-27b-it base model using Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) techniques on H100-80G GPUs.

Key Capabilities

• Multilingual Proficiency: Trained with cross-lingual data for Korean, Chinese, English, and Japanese, enhancing understanding across these languages.
• Logical Reasoning: Specifically designed to handle complex Korean logical problems.
• Enhanced Analysis: Strengthened for high-level analysis of customer reviews and social media postings.
• Coding Support: Features enhanced capabilities for coding tasks.
• Efficient Training: Utilizes Deepspeed Stage 3, rslora, and BAdam Layer Mode for efficient training.

Performance

This model achieved Rank-1 on the Open Ko LLM Leaderboard Season 2 (2024/11/01~2024/12/28) with an average score of 51.37. Notable benchmark scores include:

• Ko-GPQA: 25.25
• Ko-Winogrande: 68.27
• Ko-GSM8k: 70.96
• Ko-EQ Bench: 50.25
• Ko-IFEval: 49.84

Use Cases

This model is well-suited for applications requiring advanced Korean language understanding, complex logical problem-solving, multilingual text processing, and specialized tasks like customer feedback analysis or code generation.