BioMedGPT-Mol is an 8 billion parameter multimodal molecular language model developed by PharMolix Inc. and the Institute of AI Industry Research (AIR), Tsinghua University. It is designed for both molecular understanding and generation, supporting tasks like chemical name conversion, molecular captioning, property prediction, and molecule editing. Trained with a multi-task curriculum, this model excels across diverse molecule-centric discovery benchmarks with a 32768 token context length.
Loading preview...
BioMedGPT-Mol: Multimodal Molecular Language Model
BioMedGPT-Mol is an 8 billion parameter multimodal molecular language model developed by PharMolix Inc. and the Institute of AI Industry Research (AIR), Tsinghua University. This model is specifically engineered for advanced molecular understanding and generation tasks, leveraging a well-structured multi-task training curriculum.
Key Capabilities
- Molecular Understanding: Supports tasks such as chemical name conversion, molecular captioning, and property prediction.
- Molecular Generation: Capable of reaction modeling, molecule editing, and property optimization.
- Multimodal Functionality: Integrates diverse data types relevant to molecular science.
- High Performance: Demonstrates strong performance across various molecule-centric discovery benchmarks.
- Context Length: Features a substantial context window of 32768 tokens, enabling processing of complex molecular data.
Good For
- Drug Discovery: Accelerating early-stage drug discovery processes through advanced molecular modeling.
- Chemical Research: Assisting researchers with tasks like synthesizing new compounds or predicting molecular behavior.
- Bioinformatics: Applications requiring deep understanding and manipulation of molecular structures and properties.
For more technical details, refer to the technical report.