Fine-Tuning Small Language Models for Reliable VASP INCAR Generation
INCAR-SLM outperforms large LLMs by fine-tuning a small model (Qwen3-4B) on reference VASP calculations to reliably generate physics-sensitive INCAR files. This system separates generation from post-processing using VASPGuard, ensuring the output adheres to complex material constraints, making it practical for local, high-throughput materials workflows.