M-SOLAR-10.7B-v1.0: An Instruction-Tuned Korean LLM

M-SOLAR-10.7B-v1.0 is a 10.7 billion parameter language model developed collaboratively by Megastudy Education, Prediction, and Myes. This model has undergone extensive instruction fine-tuning using a variety of Korean datasets, emphasizing diverse linguistic tasks and domain-specific knowledge.

Key Capabilities

• Instruction Following: Trained on a range of instruction-based datasets, including kyujinpy/KOR-OpenOrca-Platypus-v3, enabling it to understand and execute complex prompts.
• Text Summarization and Generation: Utilizes AIHUB technical science summary data to reconstruct original texts from summaries and clues, demonstrating strong summarization and text expansion abilities.
• Contextual Understanding: Incorporates a unique "Random Split Generation" strategy where the model learns to infer the original order of shuffled sentences, enhancing its understanding of textual coherence and logical flow.
• Domain Adaptation: Leverages proprietary in-house educational data from Megastudy Education and Prediction, suggesting potential strengths in educational content generation and analysis.

Good For

• Korean Language Processing: Optimized for tasks requiring deep understanding and generation of Korean text.
• Instruction-Based Tasks: Excels in scenarios where precise instruction following is critical.
• Educational Applications: Its training on educational domain data makes it suitable for tasks related to learning content, question answering, and summarization in an academic context.