metadata
language: hu
license: apache-2.0
datasets:
- common_crawl
- wikipedia
tags:
- byte representation
- gradient boosting
- hungarian
Charmen-Electra
A byte-based transformer model trained on Hungarian language. In order to use the model you will need a custom Tokenizer which is available at: https://github.com/szegedai/byte-offset-tokenizer.
Since we use a custom architecture with Gradient Boosting, Down- and Up-Sampling, you have to enable Trusted Remote Code like:
model = AutoModel.from_pretrained("SzegedAI/charmen-electra", trust_remote_code=True)
Acknowledgement
