---
tags:
- molecular
- chemistry
- tokenizer
---


# MoleBERT Tokenizer

This model is a molecular tokenizer trained on Nb-based TMC. From scratch, 400k train molecules, 30 epochs
Test loss = 0.22
Code from: https://github.com/Rich-XGK/GTMGC

## Usage

```python
from models.mole_bert_tokenizer import MoleBERTTokenizerCollator, MoleBERTTokenizer
tokenizer = MoleBERTTokenizer.from_pretrained(...)