TerraMARS: A Domain-Adapted Small-Language-Model Pipeline for Mars Terraforming Literature
Researchers have developed TerraMARS, an information extraction pipeline designed to process scientific literature about Mars terraforming. This system utilizes a fine-tuned Google Gemma 3 1B small language model, adapted using QLoRA, to answer questions and convert unstructured text into machine-readable JSON format. The goal is to integrate this knowledge into applications like digital twins and habitability models for Mars, though further accuracy improvements are noted as necessary. AI
IMPACT Enables more efficient knowledge extraction from scientific literature for specialized domains like Mars terraforming.