Code and Data for Browser-Native Autonomous Web-based Geographical Information Systems Using Small Language Models

This repository contains the source code, trained models, and benchmark datasets supporting the study “Fine-Tuning Small Language Models (SLMs) for Autonomous Web-based Geographical Information Systems (AWebGIS)”. The materials enable reproduction of experiments comparing a cloud-based large language model (DeepSeek Chat V3.1) with browser-executable, fine-tuned Small Language Models (T5-small, T5-efficient-mini, and T5-efficient-tiny) for executing geospatial operations from natural language input.The codebase implements a fully decentralized AWebGIS architecture running entirely in the web browser using WebAssembly and WebGPU, eliminating reliance on centralized cloud infrastructures. The dataset includes natural language–to–geospatial-function mappings used for training and evaluation, along with scripts for performance assessment using Exact Match Accuracy, Levenshtein Similarity, and ROUGE metrics.These resources support research on privacy-preserving, cost-efficient, and installation-free autonomous Web GIS systems.