GeoGPT-Research-Project/GeoGPT-R1-Preview
GeoGPT-R1-Preview is a 72.7 billion parameter large language reasoning model developed by GeoGPT-Research-Project, built upon the Qwen2.5-72B foundation. This model is specifically enhanced for specialized geoscience applications through continual pre-training, supervised fine-tuning, and direct preference optimization. It offers remarkable reasoning capabilities for answering complex geoscience questions, making it ideal for researchers and professionals in the field. The model supports primary English and Chinese languages and has a context length of 32768 tokens.
Loading preview...
GeoGPT-R1-Preview: A Specialized Geoscience Reasoning Model
GeoGPT-R1-Preview is a 72.7 billion parameter large language model developed by the GeoGPT-Research-Project, specifically designed to advance geosciences research. Built on the Qwen2.5-72B foundation, this model undergoes a rigorous three-stage post-training process to enhance its specialized capabilities:
Key Capabilities
- Geoscience Specialization: Enhanced through Continual Pre-training (CPT) on a diverse set of geoscience-related corpora, creating a solid specialized model.
- Instruction Following: Improved via Supervised Fine-tuning (SFT) using QA pairs labeled by geoscientists and generated from the CPT corpus.
- Reasoning: Features remarkable reasoning capabilities for answering geoscience questions, aligned with human expectations through Direct Preference Optimization (DPO).
- Data Integrity: Trained on authoritative and impartial data, including a geoscience-specific subset of CommonCrawl and approximately 280,000 open-access publications with CC BY or CC BY-NC licenses.
- Multilingual Support: Primarily supports English and Chinese languages.
Good for
- Geoscience Research: Provides valuable AI tools for scientists, researchers, and professionals engaged in geoscience research worldwide.
- Non-commercial Use: Primarily intended for non-commercial research and educational purposes.
- Specialized Question Answering: Excels at answering complex questions within the geoscience domain.