jiosephlee/intern-s1-mini-context-conditioned-molecule-transfer-v10-3-0-2-bioavailability-ma-BEST
The jiosephlee/intern-s1-mini-context-conditioned-molecule-transfer-v10-3-0-2-bioavailability-ma-BEST model is an 8 billion parameter, context-conditioned molecule transfer model developed by jiosephlee, based on the Intern-S1-mini-lm architecture. It is specifically fine-tuned for predicting oral bioavailability, utilizing a mixed-continuous training artifact. This model excels in tasks related to molecular property prediction, particularly for bioavailability, as evidenced by its validation-selected performance metrics.
Loading preview...
Model Overview
This model, intern-s1-mini-context-conditioned-molecule-transfer-v10-3-0-2-bioavailability-ma-BEST, is an 8 billion parameter, context-conditioned molecule transfer model developed by jiosephlee. It is built upon the Intern-S1-mini-lm base model and has been specifically fine-tuned for predicting oral bioavailability.
Key Capabilities
- Oral Bioavailability Prediction: Optimized for tasks involving the prediction of oral bioavailability of molecules.
- Context-Conditioned Molecule Transfer: Leverages a context-conditioned approach for molecular property prediction.
- Performance Metrics: Achieves a validation Macro-F1@5 of 0.6401 and a Test Macro-F1@5 of 0.7066, indicating strong performance in its specialized domain.
Training Details
The model was trained for 10 epochs with a seed of 42, using a soft-target loss function. It was selected based on the validation knn_binary_macro_f1_at_5 metric at optimizer step 160. The training data used was the V10.3.0.2 Bioavailability Ma mixed-continuous artifact, which is a local artifact not published as a Hub dataset.
Use Cases
This model is particularly well-suited for research and development in cheminformatics and drug discovery, where accurate prediction of molecular properties like oral bioavailability is crucial. It provides a specialized tool for scientists and researchers working with molecular data.