awebgis-tiny is a fine-tuned T5-efficient-tiny model designed for Autonomous Web-based Geographical Information Systems (AWebGIS). This compact model enables natural language to geospatial function call conversion, allowing users to interact with web-based GIS applications through conversational queries. The model is optimized for deployment in resource-constrained environments while maintaining effective performance for geospatial operations.

The primary goal of this model is to bridge the gap between natural language understanding and geospatial functionality, enabling autonomous operation of web-based GIS systems without relying on cloud-based large language models, thus ensuring privacy and reducing latency.

Model Details

Base Model: google/t5-efficient-tiny
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: ~16 million
Floating Point Precision: FP32 (32-bit)
Approximate Model Size: ~64 MB

Model Variants

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

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

How to Use

python

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

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

# Example usage
input_text = "Find my location on the map!"
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:

Training Information: Check the GitHub repository for training scripts and documentation
Data Format: Prepare your dataset with natural language inputs and corresponding geospatial function call outputs
Training Setup: Follow the instructions in the repository to set up the Python environment and dependencies
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:

bibtex

@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 google/t5-efficient-tiny, 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.