lhpku20010120/Omni-Edu-27B
OmniEdu-27B by lhpku20010120 is a 27 billion parameter multimodal foundation model, fine-tuned from Qwen3.8-27B, specifically designed for K–12 learning and teaching. It integrates subject competence, curriculum grounding, diagnostic reasoning, and pedagogical scaffolding to support educational tasks. With a 32,768-token context length, this model excels at solving K–12 problems, identifying misconceptions, and providing adaptive instructional guidance.
Loading preview...
OmniEdu-27B: Open Foundation Model for K–12 Education
OmniEdu-27B is a 27 billion parameter multimodal model, fine-tuned from Qwen3.8-27B, developed by researchers at Peking University and the University of the Chinese Academy of Sciences. It is specifically engineered to support K–12 learning and teaching by connecting subject knowledge, curriculum understanding, learner diagnosis, and instructional support. This model represents the largest and most capable checkpoint within the OmniEdu family, demonstrating strong performance across various educational benchmarks.
Key Capabilities
- Subject Competence: Solves K–12 problems across diverse subjects and input formats.
- Curriculum Grounding: Links educational content to curriculum standards and prerequisites.
- Diagnostic Reasoning: Identifies learner misconceptions, missing prerequisites, and knowledge gaps.
- Pedagogical Action: Provides targeted feedback, asks guiding questions, explains concepts, and adapts instructional support.
- Multimodal Support: Handles both text and image-based questions, making it versatile for various educational materials.
Good For
- Educational Research: A robust foundation for exploring AI applications in K–12 education.
- Teacher-Assistance Tools: Developing tools that aid educators in lesson planning, assessment, and student support.
- Learning-Support Prototypes: Creating innovative prototypes for personalized learning experiences and adaptive tutoring systems.
- High-Performance Educational Tasks: Achieving top results on benchmarks like K12-Bench, MathFish, EDUMATH MaC, and LongTutor Teaching, as detailed in the accompanying paper.
OmniEdu-27B was trained using full-parameter supervised fine-tuning on a mixture of 69,999 instruction examples, including over 60,000 education-specific examples, with a 32,768-token training sequence length. It is intended for educational research and development, with a recommendation for educators to review generated material before use.