zjunlp/MolGen-7b
MolGen-7b by zjunlp is a 7 billion parameter molecular generative model built using the LlamaForCausalLM architecture. It is specifically designed for generating novel molecules and completing partial molecular structures, leveraging molecular language SELFIES. This model excels at de novo molecule generation and molecular completion tasks, offering a specialized tool for chemical research and drug discovery.
Loading preview...
Overview
zjunlp/MolGen-7b is a 7 billion parameter molecular generative model based on the LlamaForCausalLM architecture. It is specifically trained to understand and generate molecular structures using SELFIES (Simplified Molecular-input Line-entry System) as its molecular language. This model provides a specialized tool for tasks within chemistry and drug discovery.
Key Capabilities
- De Novo Molecule Generation: The model can generate entirely new molecular structures from a starting token, enabling the exploration of novel chemical spaces.
- Molecular Completion: Users can input partial molecular structures, and the model will complete them, assisting in the design and optimization of compounds.
- SELFIES-based Generation: By utilizing SELFIES, the model ensures that generated molecules are chemically valid and syntactically correct, which is a common challenge with other molecular representations.
Intended Uses
This model is ideal for researchers and developers in chemistry, materials science, and pharmaceuticals who need to:
- Discover new chemical entities.
- Optimize existing molecular scaffolds.
- Explore diverse molecular libraries for specific properties.
Citation
If you use this model, please cite the associated research paper: Domain-Agnostic Molecular Generation with Chemical Feedback.