---
license: cc-by-4.0
language:
- en

tags:
- emergency
- disaster-preparedness
- fema
- cdc
- usgs
- dhs
- public-safety
- instruction-tuning
- safety
- crisis-response
- multi-hazard

task_categories:
- text-generation
- question-answering

pretty_name: Emergency Response Instructions
size_categories:
- 10K<n<100K
---

# Emergency Response Instructions

A supervised fine-tuning (SFT) dataset built from official government and international organization documents focused on disaster preparedness, emergency response, and crisis safety.

The dataset consolidates trusted guidance from agencies like FEMA, CDC, USGS, DHS, WHO, IFRC, UNICEF, Red Cross, and more — transforming them into structured instruction-following examples.

---

## Coverage

This dataset spans multi-hazard scenarios, including:

- Earthquakes
- Floods
- Hurricanes & storms
- Wildfires
- Tornadoes
- Extreme heat & winter storms
- Power outages
- Landslides & tsunamis
- Volcanoes & ash hazards
- Nuclear & radiological events
- Chemical incidents
- Cybersecurity emergencies
- Water safety & sanitation
- Public health emergencies
- War & civil defense preparedness

---

## Instruction Design

Each document is chunked and transformed into 7 instruction archetypes:

1. Summary
2. Step-by-step checklist
3. Key facts
4. Before / During / After
5. Explain-why reasoning
6. Emergency supplies list
7. Yes/No + reasoning

---

## Dataset Characteristics

- Source-backed, high-trust content
- Instruction-tuned for real-world usability
- Multi-format reasoning
- Designed for safety-critical AI systems
- Suitable for:
  - Emergency assistants
  - Disaster response copilots
  - Public safety chatbots
  - Educational tools

---

## Sources

### U.S. Government Agencies
- FEMA (ready.gov, fema.gov)
- CDC (Centers for Disease Control and Prevention)
- USGS (United States Geological Survey)
- DHS (Department of Homeland Security)
- EPA (Environmental Protection Agency)
- NOAA / NWS (Weather services)

### International Organizations
- WHO (World Health Organization)
- IFRC (International Federation of Red Cross)
- UNICEF
- Red Cross
- Save the Children
- CARE
- Mercy Corps
- World Vision

### Government & Policy Frameworks
- UK National Risk Register (2025)
- Sweden Civil Preparedness Guide
- Australia Crisis Framework
- Canada Preparedness Guide
- NATO Civil Preparedness

---

## Source Documents

Includes 60+ verified PDFs such as FEMA Are You Ready Guide, CDC Emergency Supply Kit Guide, Ready.gov hazard sheets, USGS handbooks, nuclear response guides, and international preparedness frameworks.

Full mapping of dataset files to URLs is included in the dataset as SOURCE_URLS.

---

## Using with Transformers

Example: fine-tune a causal language model using Hugging Face Transformers.

```py
from datasets import load_dataset

ds = load_dataset("nuhmanpk/emergency-response-instructions")
```

or

```py
from datasets import load_dataset
from transformers import AutoModelForCausalLM, AutoTokenizer, DataCollatorForLanguageModeling, Trainer, TrainingArguments

model_name = "distilgpt2"
max_length = 1024

dataset = load_dataset("json", data_files={"train": "train.jsonl", "validation": "validation.jsonl"})

tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.pad_token = tokenizer.eos_token

def tokenize_function(examples):
    return tokenizer(examples["text"], truncation=True, max_length=max_length)

tokenized_dataset = dataset.map(tokenize_function, batched=True, remove_columns=dataset["train"].column_names)

model = AutoModelForCausalLM.from_pretrained(model_name)

data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)

training_args = TrainingArguments(
    output_dir="emergency-response-instructions-model",
    eval_strategy="epoch",
    learning_rate=2e-5,
    weight_decay=0.01,
    num_train_epochs=3,
    push_to_hub=False,
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_dataset["train"],
    eval_dataset=tokenized_dataset["validation"],
    data_collator=data_collator,
    processing_class=tokenizer,
)

trainer.train()

```

---

## License

- Source documents: public domain or respective organizational ownership
- Dataset compilation: CC-BY-4.0

---

## Disclaimer

This dataset is derived from official guidance but should not replace real-time instructions from local authorities during an actual emergency.

---

## Use Cases

- Fine-tuning safety-aware LLMs
- Emergency preparedness education
- Crisis simulation systems
- AI copilots for disaster response
- Public safety automation tools

---

## Author

- GitHub: https://github.com/nuhmanpk  
- Kaggle: https://www.kaggle.com/nuhmanpk  
- LinkedIn: https://www.linkedin.com/in/nuhmanpk