MiaLatte-Indo-Mistral-7b: Indonesian Language Model

MiaLatte-Indo-Mistral-7b is a 7 billion parameter language model developed by indischepartij, specifically optimized for the Indonesian language. It is a derivative of the OpenMia model, enhanced through a merge of three distinct models using the MergeKit's dare_ties method.

Key Capabilities

• Indonesian Language Proficiency: Designed to answer everyday questions specifically in Bahasa Indonesia.
• Merged Architecture: Combines indischepartij/OpenMia-Indo-Mistral-7b-v2, Obrolin/Kesehatan-7B-v0.1, and FelixChao/WestSeverus-7B-DPO-v2 to leverage their respective strengths.
• Performance: Achieves an average score of 67.86 on the Open LLM Benchmark, including 66.55 on AI2 Reasoning Challenge and 63.93 on MMLU.

Good For

• Applications requiring robust natural language understanding and generation in Bahasa Indonesia.
• Building chatbots or conversational agents for Indonesian-speaking users.
• Tasks involving general knowledge and everyday queries in an Indonesian context.