Hugging Face Overview
Linkbricks Horizon-AI Korean Llama 3.1 (70B)
This model, developed by Yunsung Ji (Saxo), a data scientist at Linkbricks Horizon-AI, is a 70 billion parameter Korean language model. It is fine-tuned from the NousResearch/Meta-Llama-3.1-70B-Instruct base model using Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) on KT-CLOUD's H100-80G GPUs.
Key Capabilities
- Multilingual Enhancement: Trained with Korean-Chinese-English-Japanese cross-training data, improving cross-lingual understanding.
- Logical Reasoning: Enhanced to handle complex Korean logical problems through specialized logical data training.
- Domain-Specific Strengths: Particularly strong in high-level analysis of customer reviews and social media postings, as well as coding tasks.
- Tool Calling: Supports tool calling functionalities.
- Context Window: Utilizes a 32768-token context window.
- Tokenizer: Retains the base model's tokenizer without word expansion.
Good For
- Applications requiring advanced Korean language understanding and generation.
- Tasks involving multilingual text analysis, especially across Korean, Chinese, English, and Japanese.
- Use cases demanding logical problem-solving in Korean.
- Analyzing customer feedback, social media trends, and coding assistance.