aakothari/Conformer

Conformer

A browser workbench for the BRAID molecular line notation and the BRAIDBERTa / DeepBERTa encoders. Load molecules, convert between representations, draw structures, compute descriptors, benchmark tokenisations, embed and compare encoders, export datasets.

Nothing to install. Structure parsing, drawing, descriptors and fingerprints run entirely in your browser via RDKit's WebAssembly build, so most of the workbench works with no server at all.

BRAID conversion and the encoders need Python — braids.codec depends on RDKit APIs that the WebAssembly build does not expose — so those routes are served by a small FastAPI service:

https://braid-9wc2.onrender.com

The encoders run there from ONNX graphs exported from the checkpoints below, on CPU. That service is on a free instance and sleeps when idle: the first request after a quiet spell can take ~50 seconds to wake it. Everything browser-side stays instant meanwhile.

• —Code: https://github.com/AayushK-othari/braid
• —Models: https://huggingface.co/aakothari/BRAIDBERTa and https://huggingface.co/aakothari/DeepBERTazincbase100kv4

GET /version on the backend returns the git SHA, the pinned model revisions and the installed package versions behind whatever you are looking at.

Pointing at your own backend

The workbench's settings panel takes any endpoint. To run the converters and encoders locally instead — nothing leaves your machine:

pip install -r requirements-backend.txt
python backend.py

then set the endpoint to http://127.0.0.1:8000.