CrystalReasoner/Qwen2.5-3B-CrysReas-NoValidityTerm
CrystalReasoner/Qwen2.5-3B-CrysReas-NoValidityTerm is an end-to-end LLM framework developed by CrystalReasoner, based on the Qwen2.5-3B architecture. This model is specifically designed for generating crystal structures from natural language instructions. It utilizes supervised fine-tuning, incorporates crystallographic and physical priors through thinking traces, and employs reinforcement learning with verifiable rewards to enhance the validity, stability, and property conditioning of generated structures. Its primary strength lies in its ability to translate natural language descriptions of bulk materials into detailed crystal structure parameters, including lattice vectors and atomic coordinates.
Loading preview...
CrystalReasoner/Qwen2.5-3B-CrysReas-NoValidityTerm Overview
This model, developed by CrystalReasoner, is an end-to-end Large Language Model (LLM) framework built upon the Qwen2.5-3B architecture. Its core function is to generate crystal structures directly from natural language instructions, making it a specialized tool for materials science and chemistry applications.
Key Capabilities
- Property-Conditioned Crystal Generation: Generates crystal structures based on natural language descriptions that include chemical formulas and physical properties (e.g., bulk modulus).
- Reasoning Integration: Employs "thinking traces" to incorporate crystallographic and physical priors before generating atomic coordinates, enhancing the scientific accuracy of the output.
- Reinforcement Learning (RL): Utilizes RL with verifiable rewards to improve the validity, stability, and property conditioning of the generated structures.
- Structured Output: Capable of outputting detailed descriptions of lattice vectors, angles, element types, and atomic coordinates, which can then be converted into
pymatgenStructure format for further analysis.
Good For
- Materials Scientists: Automating the generation of hypothetical crystal structures based on desired properties.
- Researchers: Exploring new material compositions and structures through natural language prompts.
- Computational Chemists: Streamlining the initial stages of crystal structure prediction and design.
This model is particularly useful for tasks requiring the translation of high-level material descriptions into precise, physically plausible crystal structure data.