scent-to-molecule

🧬 Scent to Molecule Controt

A ControlNet-style model that generates molecular structures (SMILES) from scent descriptions.

Model Description

This model converts natural language scent descriptions into chemically valid SMILES representations of molecules that would produce those scents.

Model Details

• Training epochs: 20
• Best validation loss: 0.030441686697304248
• Model size: 2.9 MB
• Architecture: ControlNet-style adapter with frozen molecular backbone
• Text encoder: sentence-transformers/all-MiniLM-L6-v2

Usage

from huggingface_hub import hf_hub_download
import torch

# Download model
model_path = hf_hub_download("munchers/scent-to-molecule", "best_control.pt")

model.load_state_dict(checkpoint['model_state_dict'])
if torch.cuda.is_available():
    model = model.cuda()

Examples

Training Data

• Training samples: 815 compounds
• Validation samples: 157 compounds
• Chemical categories: 8 (esters, aldehydes, terpenes, phenolics, etc.)

Limitations

• Uses mock molecular backbone (not full physics simulation)
• Template-based SMILES generation
• English-only descriptions
• Synthetic training dataset

Citation

@misc{scent-to-molecule-controlnet,
  title={Scent-to-Molecule Control},
  author={Shiva Mudide},
  year={2025},
  howpublished={\url{https://huggingface.co/munchers/scent-to-molecule}}
}