Parks LLM v2 - US Parks Assistant 🏞️

A conversational AI model fine-tuned to help with US national parks, state parks, and recreation areas. This is an improved version with better training data and fixed attention mask handling.

Model Description

Parks LLM v2 is fine-tuned from microsoft/DialoGPT-medium on a curated dataset of park-related conversations. It provides helpful information about:

🎒 Packing and Preparation: What to bring for different park environments
📅 Visit Planning: Best times to visit, crowd levels, weather considerations
🏕️ Camping & Reservations: Booking guidance and accommodation options
👨‍👩‍👧‍👦 Family Travel: Family-friendly park recommendations and activities
💰 Budget Tips: Cost-saving strategies and fee information
🐻 Safety Guidance: Wildlife safety and Leave No Trace principles
🥾 Activities: Hiking, camping, and outdoor activity recommendations

Quick Start

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model
tokenizer = AutoTokenizer.from_pretrained("ajc2195/parks-llm-v2")
model = AutoModelForCausalLM.from_pretrained("ajc2195/parks-llm-v2")

def ask_parks_assistant(question):
    prompt = f"Human: {question}\nAssistant:"
    inputs = tokenizer(prompt, return_tensors="pt", padding=True)
    
    with torch.no_grad():
        outputs = model.generate(
            inputs['input_ids'],
            attention_mask=inputs['attention_mask'],
            max_new_tokens=80,
            temperature=0.7,
            do_sample=True,
            repetition_penalty=1.2,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id
        )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("Assistant:")[-1].strip()

# Example usage
answer = ask_parks_assistant("What should I pack for desert parks?")
print(answer)

Pipeline Usage

from transformers import pipeline

parks_assistant = pipeline(
    "text-generation", 
    model="ajc2195/parks-llm-v2",
    tokenizer="ajc2195/parks-llm-v2"
)

response = parks_assistant(
    "Human: What's the best time to visit Yellowstone?\nAssistant:",
    max_new_tokens=60,
    temperature=0.7,
    do_sample=True,
    repetition_penalty=1.2
)

Example Conversations

Q: What should I pack for Yellowstone? A: For Yellowstone, pack layers for changing weather, waterproof jacket, hiking boots, bear spray, sunscreen, hat, and warm clothes for evenings. Don't forget your camera and binoculars for wildlife viewing!

Q: When is the best time to visit Zion? A: The best time to visit Zion is spring (March-May) and fall (September-November) when temperatures are mild and crowds are smaller. Summer can be very hot, while winter may have snow at higher elevations.

Q: How do I make campground reservations? A: Make reservations at recreation.gov up to 5 months in advance. Popular campgrounds fill quickly, so book early. Some sites are first-come, first-served, arriving early in the morning is recommended.

Model Details

Base Model: microsoft/DialoGPT-medium
Training Data: 15 high-quality park conversation examples
Training Method: Fine-tuning with improved hyperparameters
Specialization: US National Parks and recreation areas
Version: 2.0 (improved from original parks-llm)

Improvements in v2

✅ Fixed attention mask warnings
✅ Better training data diversity
✅ Improved repetition penalty
✅ Proper pad token handling
✅ Enhanced generation parameters

Usage Tips

Use the conversation format: "Human: [question]\nAssistant:"
Include attention_mask in generation for best results
Set repetition_penalty=1.2 to avoid repetitive outputs
Recommended temperature=0.7 for good balance of coherence and variety

Limitations

Focuses primarily on US parks and recreation areas
Provides general guidance - always verify with official park sources
Not a substitute for real-time park information, alerts, or conditions
Training data is limited - may not cover all park scenarios

Ethical Considerations

This model should be used as a helpful assistant, not a replacement for official park information
Always check current park conditions, alerts, and regulations before traveling
Be aware that park policies and conditions change frequently

License

Apache 2.0

Citation

If you use this model, please cite:

@misc{parks-llm-v2,
  title={Parks LLM v2: Conversational AI for US Parks and Recreation},
  author={ajc2195},
  year={2025},
  url={https://huggingface.co/ajc2195/parks-llm-v2}
}

Built with ❤️ for park enthusiasts and outdoor adventurers!

ajc2195
/

parks-llm-v2