CrystalReasoner/Qwen2.5-3B-CrysReas
CrystalReasoner/Qwen2.5-3B-CrysReas is an end-to-end LLM framework developed by CrystalReasoner for generating crystal structures from natural language instructions. This model utilizes supervised fine-tuning (SFT) for crystal-structure generation and incorporates thinking traces to integrate crystallographic and physical priors. It further employs reinforcement learning (RL) with verifiable rewards to enhance the validity, stability, and property conditioning of generated structures. Its primary application is the property-conditioned generation of crystal structures based on textual descriptions.
Loading preview...
CrystalReasoner/Qwen2.5-3B-CrysReas Overview
CrystalReasoner/Qwen2.5-3B-CrysReas is an innovative language model framework designed for the generation of crystal structures from natural language instructions. Developed by CrystalReasoner, this model stands out by integrating advanced techniques to ensure the accuracy and utility of its outputs.
Key Capabilities
- End-to-end Crystal Structure Generation: Directly translates natural language descriptions into crystal structures.
- Supervised Fine-Tuning (SFT): Employs SFT to teach the model the fundamentals of crystal structure generation.
- Incorporation of Physical Priors: Utilizes "thinking traces" to introduce crystallographic and physical knowledge before generating atomic coordinates, enhancing structural realism.
- Reinforcement Learning (RL): Leverages RL with verifiable rewards to significantly improve the validity, stability, and property conditioning of the generated crystal structures.
- Pymatgen Integration: Provides a script to convert generated text into
pymatgen.core.Structureobjects, facilitating downstream materials science analysis.
Good For
- Researchers and engineers needing to generate novel crystal structures based on desired properties or chemical formulas.
- Automating the initial design phase in materials discovery and engineering.
- Exploring the relationship between natural language descriptions and complex material structures.