🌏 Environmental Issues Extractor

📑 Project Description

Supervised fine-tuned LLM that detects and quantifies environment issues from 10,000+ environmental article titles.

📦 Distillation LLMs

👩🏻‍🏫 Selected Teacher: Qwen/Qwen3-4B-Instruct-2507

https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507

Send 10,000+ environmental article titles to the teacher LLM, letting it extract and quantify environmental issues and save all results.

👨🏻‍🎓 Selected Student: Qwen/Qwen2.5-3B-Instruct

https://huggingface.co/Qwen/Qwen2.5-3B-Instruct

Split 80% of the teacher's extracted and quantified issues to fine-tune the student LLM, and 20% for the evaluation on student LLM.

📊 Evaluation Metrics

MAE: 0.004736221986196294

RMSE: 0.035369152172009685

Total samples: 2187

📲 Example Issues Extraction

{
  "title": "Carbon emissions from England's roads plan '100 times greater than government claims'",
  "issues": {
    "air pollution": 1,
    "biodiversity loss": 1,
    "climate change": 9,
    "deforestation": 1,
    "energy crisis": 1,
    "fossil fuel dependence": 10,
    "waste": 1
  }
}