CrystalReasoner/Qwen2.5-3B-CrysReas-ElasticProperties

CrystalReasoner/Qwen2.5-3B-CrysReas-ElasticProperties Overview

This model is a specialized 3 billion parameter Qwen2.5-based language model developed by CrystalReasoner, designed for generating crystal structures from natural language descriptions. It is part of the broader CrystalReasoner (CrysReas) framework, which focuses on reasoning and reinforcement learning for property-conditioned crystal structure generation.

Key Capabilities

• Crystal Structure Generation: Generates detailed crystal structures, including lattice vectors, element types, and atomic coordinates, from natural language prompts.
• Property Conditioning: Capable of generating structures conditioned on specific material properties, such as bulk modulus, as demonstrated in the example for elastic properties.
• Reasoning Integration: Incorporates "thinking traces" to apply crystallographic and physical priors before generating atomic coordinates.
• Reinforcement Learning (RL): Utilizes RL with verifiable rewards to enhance the validity, stability, and property conditioning of the generated structures.
• Pymatgen Compatibility: Generated structures can be easily converted into pymatgen Structure format for further analysis and manipulation.

Good For

• Researchers and engineers in materials science needing to generate crystal structures based on desired chemical formulas and physical properties.
• Automated design and discovery of new materials with specific elastic characteristics.
• Applications requiring the programmatic generation of crystallographic data from high-level descriptions.