# Model Card: WorldDisasterLM

## Model Details

- **Model Name:** WorldDisasterLM
- **Alternative Names:** DisasterGPT, CrisisMind, OpenDisasterAI, GlobalRescueLM, HumanitarianGPT
- **Base Model:** meta-llama/Llama-3.1-8B-Instruct
- **Architecture:** Decoder-only transformer, instruction tuned
- **Future Upgrades:** 70B checkpoints, MoE variants
- **Primary Domains:** Disaster management, emergency response, humanitarian aid, risk analytics

## Intended Use

### Primary Users

- Government agencies
- NGOs and humanitarian organizations
- Emergency responders
- Researchers and policy groups
- Healthcare organizations
- Citizens and volunteers

### Intended Tasks

- Disaster Q&A
- Emergency guidance generation
- Incident classification
- Risk scoring by region/event
- Resource planning recommendations
- Situation report summarization

## Training Data

Aggregated disaster corpora from international organizations, open disaster databases, research literature, and near-real-time alert metadata. Data is normalized into instruction-friendly samples and multilingual pairs.

## Evaluation

Core metrics include:

- Response accuracy
- Hallucination rate
- Safety policy compliance
- Emergency-response correctness
- Multilingual performance across 10 target languages

## Safety and Risk

- Not a replacement for emergency command centers
- Outputs should be verified with authoritative real-time sources
- Critical instructions must involve human oversight
- High-risk outputs are tagged for escalation

## Limitations

- Data availability and timeliness may vary by region
- Some low-resource languages may have lower response quality
- Unknown edge-case events may reduce reliability

## License

MIT