KT has announced the release of its open-source large language model (LLM), “Mi:dm 2.0,” on the Hugging Face platform to foster the adoption of Korean-centric AI. Launched in 2023, Mi:dm is designed specifically for the Korean language and context. The Mi:dm 2.0 suite comprises two versions: “Mi:dm 2.0 Base,” featuring 11.5 billion parameters, and the lightweight “Mi:dm 2.0 Mini” with 2.3 billion parameters, both supporting Korean and English. Trained on a diverse Korean-specific dataset, the model emphasizes content compliance and ethical AI use, with a custom tokenizer to enhance understanding of local nuances. Collaborating with Korea University, KT ensured the model aligns with responsible AI principles and international guidelines. By providing a powerful domestic alternative, KT aims to enhance its global competitiveness while planning further collaborations with Microsoft to fine-tune GPT-4 according to Korean cognitive patterns.
Source link
Share
Read more