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

# BERTose Glycan Encoder

This repository contains the BERTose checkpoint for WURCS glycan embedding inference. It is the release-facing glycan representation model used by the companion notebook.

## Quick Start

The recommended user path is the companion notebook:

```python
from huggingface_hub import hf_hub_download

checkpoint = hf_hub_download(
    repo_id="supanthadey1/bertose-glycan-encoder",
    filename="checkpoints/bertose_glycan_encoder.pt",
)
vocab = hf_hub_download(
    repo_id="supanthadey1/bertose-glycan-encoder",
    filename="vocab/bpe_vocabulary.json",
)
```

No Hugging Face token is required for this BERTose checkpoint now that the repository is public.

## Files

- `checkpoints/bertose_glycan_encoder.pt` - BERTose glycan encoder checkpoint.
- `vocab/bpe_vocabulary.json` - WURCS BPE vocabulary.
- `src/bertose_model.py` - BERTose model definition.
- `src/bertose_layers.py` - Transformer layers used by BERTose.
- `src/wurcs_bpe_tokenizer.py` - WURCS BPE tokenizer.

## Input

Provide one WURCS glycan string or a CSV batch with `sample_id,wurcs`.

Free-text glycan names, common names, SNFG drawings, and IUPAC-condensed strings are not parsed directly by this checkpoint. Convert those inputs to WURCS first, then run BERTose embedding inference.

## Output

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

## Notes

This repository does not perform IUPAC-condensed/name-to-WURCS conversion. For now, provide WURCS directly.

License metadata is currently `other`; update it when the final release license and citation text are chosen.