---
library_name: pytorch
license: other
tags:
- glycans
- wurcs
- bertose
- embeddings
- pytorch
---

# Bertose V5B1 Glycan Encoder

Draft private release for Bertose glycan embedding inference.

This repository contains the leakage-proof Bertose V5B1 checkpoint used for WURCS glycan embeddings. It is intended for inference through the companion cloud notebook.

## Files

- `checkpoints/bertose_v5b_excluded.pt` - Bertose V5B1 checkpoint for glycan embeddings.
- `vocab/bpe_vocabulary.json` - WURCS BPE vocabulary.
- `src/multimodal_glycan_bert_v3.py` - model definition.
- `src/wurcs_bpe_tokenizer.py` - WURCS BPE tokenizer.

## Expected Input

Single glycan or batch CSV with WURCS strings.

## Output

Dense glycan embeddings. The companion notebook defaults to `[CLS]` pooling and also supports mean pooling over valid glycan tokens.

## Draft Notes

This model card is intentionally minimal until the final release license, citation, and manuscript wording are locked.