GAGIS-Lab/CoastalGPT-9B
CoastalGPT-9B by GAGIS-Lab is a 9 billion parameter large language model, fine-tuned from Qwen3.5-9B, specifically designed for the coastal zone domain. It integrates data, models, and knowledge to support relational organization, process reasoning, and coordinated interaction in coastal science research. The model excels at understanding specialized coastal concepts, scientific literature, and resource metadata, aiming to enhance scientific reasoning and automated research support in this domain.
Loading preview...
CoastalGPT-9B: A Specialized LLM for Coastal Science
CoastalGPT-9B, developed by GAGIS-Lab, is a 9 billion parameter large language model fine-tuned from Qwen3.5-9B. Its core purpose is to unify and reason over data, models, and knowledge within the coastal zone domain. The project, guided by Professor Zhaoyuan Yu, Professor Linwang Yuan, and Professor Wen Luo, focuses on building a semantic representation and Chain-of-Thought (CoT) reasoning framework to connect scientific questions with research workflows.
Key Capabilities & Differentiators
- Domain-Specific Adaptation: Fine-tuned using over 100,000 coastal-domain CoT instruction samples, including basic knowledge, scientific literature metadata (400,000+ records), and coastal data/model resource metadata.
- Enhanced Scientific Reasoning: Learns analytical pathways and scientific explanation logic for coastal problems through CoT instruction data generated by Qwen3.5-27B.
- Comprehensive Coastal Understanding: Improves understanding of specialized coastal concepts, literature, data-resource metadata extraction, and model-resource explanation.
- Strong Performance on Coastal Benchmarks: Achieves 73.4% overall accuracy on a specialized coastal multiple-choice benchmark, significantly outperforming OceanGPT-basic-30B-A3B-Thinking (58.6%) and Qwen3.5-9B (57.9%).
- High Accuracy in Specific Areas: Particularly strong in "Spatial Distribution" and "Impacts & Risks" (82.0% accuracy), and "Coastal Processes" (80.0%).
Ideal Use Cases
CoastalGPT-9B is designed to support various tasks in coastal science, including:
- Metadata extraction and structured description of coastal data resources.
- Analysis and explanation of coastal model resources and their applicable scenarios.
- Explanation of modeling methods, indicator systems, and data-processing workflows.
- Analysis of scientific research questions and organization of research directions.
- Construction of intelligent service systems for coastal data-model-knowledge collaboration.
- Automated research workflow support for coastal science.