awebgis-small

awebgis-small

Model Description

awebgis-small is a fine-tuned T5-small model engineered for Autonomous Web-based Geographical Information Systems (AWebGIS). As the most capable variant in the AWebGIS model family, it delivers superior performance for complex geospatial queries while maintaining efficiency suitable for web deployment. The model transforms natural language instructions into precise geospatial function calls, enabling sophisticated autonomous GIS operations.

This model is optimized for handling intricate geospatial tasks that require deeper understanding and more nuanced processing, making it ideal for applications that demand higher accuracy and can accommodate slightly larger model sizes.

Model Details

• Base Model: t5-small
• Fine-tuned for: Natural language to geospatial function call conversion
• Task: Sequence-to-sequence text generation
• Language: English
• Training Framework: PyTorch with Hugging Face Transformers

Model Specifications

• Number of Parameters: ~60 million
• Floating Point Precision: FP32 (32-bit)
• Approximate Model Size: ~240 MB

Model Variants

This model is part of a family of fine-tuned models for AWebGIS:

• awebgis-tiny - Lightweight variant for resource-constrained environments
• awebgis-mini - Balanced variant with improved accuracy
• awebgis-small (this model) - Higher capacity variant for complex queries

How to Use

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("mahdin75/awebgis-small")
model = AutoModelForSeq2SeqLM.from_pretrained("mahdin75/awebgis-small")

# Example usage
input_text = "Create a buffer zone of 10km around all hospitals, then find intersections with residential areas and calculate population density"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_length=128)
result = tokenizer.decode(outputs[0], skip_special_tokens=True)

Fine-Tuning

To fine-tune this model for your specific geospatial tasks, refer to the training documentation in the main repository:

1. Training Information: Check the GitHub repository for training scripts and documentation
2. Data Format: Prepare your dataset with natural language inputs and corresponding geospatial function call outputs
3. Training Setup: Follow the instructions in the repository to set up the Python environment and dependencies
4. Hyperparameters: Adjust learning rate, batch size, and training epochs based on your dataset size and requirements

For detailed fine-tuning instructions and training code, visit the main repository.

Live Demo

You can interact with this model through the web application:

🔗 AWebGIS Application - React-based frontend that demonstrates the model's capabilities in a real-world GIS interface

Citation

If you use this model in your research or applications, please cite the following paper:

@misc{ashani2025finetuningsmalllanguagemodels,
      title={Fine-Tuning Small Language Models (SLMs) for Autonomous Web-based Geographical Information Systems (AWebGIS)},
      author={Mahdi Nazari Ashani and Ali Asghar Alesheikh and Saba Kazemi and Kimya Kheirkhah and Yasin Mohammadi and Fatemeh Rezaie and Amir Mahdi Manafi and Hedieh Zarkesh},
      year={2025},
      eprint={2508.04846},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2508.04846},
}

Resources

• 📄 Preprint: arXiv:2508.04846
• 💻 GitHub Repository: github.com/mahdin75/awebgis
• 🌐 Web Application: github.com/mahdin75/awebgis-app

License

This model is released under the MIT License. See the LICENSE file for more details.

Important: This model is fine-tuned from t5-small, which is licensed under the Apache License 2.0. The LICENSE file includes both licenses and proper attribution. Users must comply with both license terms when using this model.