Erasmus-AI/climategpt-3-8b
ClimateGPT-3-8B is an 8 billion parameter causal language model developed by Erasmus-AI, domain-adapted from Qwen3-8B for climate science and the Planetary Boundaries framework. It features a 32768 token context length and is specifically optimized for climate and sustainability-related question answering, analysis, and content generation. The model excels in understanding and generating text within the climate domain, as evidenced by its performance on specialized Planetary Boundaries evaluation suites.
Loading preview...
Overview
ClimateGPT-3-8B is an 8 billion parameter causal language model, built upon the Qwen/Qwen3-8B base model and developed by Erasmus-AI. It has been extensively domain-adapted for climate science, with a particular focus on the Planetary Boundaries framework. The model underwent continued pretraining on climate-focused corpora, primarily derived from FineWeb-Edu, followed by supervised fine-tuning using a mix of climate instruction-following data, multi-turn conversations, and synthetic climate/Planetary Boundaries Q&A.
Key Capabilities
- Climate and Sustainability Q&A: Specialized in answering questions related to climate science and sustainability.
- Planetary Boundaries Analysis: Designed for education and analysis within the Planetary Boundaries framework.
- Climate Content Generation: Capable of drafting and summarizing climate-related content.
Performance
Evaluations using a Planetary Boundaries suite show ClimateGPT-3-8B outperforming its base model, Qwen3-8B, across various metrics. For instance, on planetary_boundaries_qa_strict_core_nolist, ClimateGPT achieved an exact match score of 0.6556 compared to Qwen3-8B's 0.4889.
Good For
- Researchers and educators in climate science.
- Applications requiring deep understanding of environmental sustainability and Planetary Boundaries.
- Generating specialized climate reports or summaries.