Skip to content

acceleratedscience/openad-service-fm4m

 
 

Repository files navigation

SMI-TED  /  Property Prediction on SMILES Input

License MIT Code style: black Docs
Linux macOS Python


About

This OpenAD service provides access to SMILES-based Transformer Encoder-Decoder (SMI-TED), a foundation model for materials science and chemistry. SMI-TED is an encoder-decoder model pre-trained on a curated dataset of 91 million SMILES samples sourced from PubChem, equivalent to 4 billion molecular tokens. SMI-TED offers several predictive models, including quantum property prediction, with two main variants ( 289 M and 8 × 289 M ).

More information:
github.com/IBM/materials
huggingface.co/ibm/materials.smi-ted
arxiv.org/abs/2407.20267

For instructions on how to deploy and use this service in OpenAD, please refer to the OpenAD docs.

Note

The current public version of app.py includes only QM9 properties from MoleculeNet. You can augment this to include QM8 and additional MoleculeNet properties (grouped under MoleculeNet) by changing the selected algorithms list, if the model checkpoints are available.


Deployment Options

About

Open-Source Wrapper for inference services with fm4m models: smi-ted and selfies-ted

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Python 91.6%
  • Smarty 3.7%
  • Dockerfile 3.3%
  • Shell 1.4%