TerraMARS: A Domain-Adapted Small-Language-Model Pipeline for Mars Terraforming Literature

TerraMARS is a newly developed information extraction pipeline that utilizes a domain-adapted Small Language Model, specifically Google Gemma 3 with 1 billion parameters, fine-tuned using Quantized Low-Rank Adaptation (QLoRA) on Mars-related datasets. This pipeline converts unstructured scientific literature on Mars into structured JSON outputs, facilitating the integration of critical planetary data into applications like digital twins and habitability modeling. Its significance lies in enhancing the extraction accuracy and factual consistency of Mars terraforming information, which is essential for future studies in planetary habitability.