CrystalReasoner/Qwen2.5-3B-CrysReas-Thinking
CrystalReasoner/Qwen2.5-3B-CrysReas-Thinking is a Qwen2.5-based language model developed by CrystalReasoner, fine-tuned for generating crystal structures from natural language instructions. This model utilizes supervised fine-tuning (SFT) and reinforcement learning (RL) to incorporate crystallographic and physical priors, improving the validity, stability, and property conditioning of generated structures. It is specifically designed for end-to-end crystal structure generation, including lattice vectors, element types, and atomic coordinates. The model's unique approach focuses on reasoning traces to enhance the physical accuracy of its outputs.
Loading preview...
CrystalReasoner/Qwen2.5-3B-CrysReas-Thinking Overview
This model, developed by CrystalReasoner, is an end-to-end language model framework built upon the Qwen2.5 architecture, specifically designed for generating crystal structures from natural language instructions. It stands out by integrating advanced techniques to ensure the physical and crystallographic accuracy of its outputs.
Key Capabilities
- Crystal Structure Generation: Translates natural language descriptions of bulk materials (e.g., chemical formula, bulk modulus) into detailed crystal structure parameters, including lattice vectors, angles, element types, and atomic coordinates.
- Reasoning Traces: Employs "thinking traces" during generation to embed crystallographic and physical priors, enhancing the scientific validity of the predicted structures.
- Reinforcement Learning (RL): Utilizes RL with verifiable rewards to further improve the validity, stability, and property conditioning of the generated crystal structures.
- Supervised Fine-Tuning (SFT): Leverages SFT to teach the model the fundamental principles of crystal structure generation.
Good For
- Researchers and engineers in materials science needing to generate novel crystal structures based on desired properties or compositions.
- Automating the initial design phase for materials with specific physical characteristics.
- Applications requiring physically informed and stable crystal structure predictions from textual input.