📝 Try out Cedille in our playground!
Cedille is available on 🤗HuggingFace
This is the most convenient way to use the model for most people.
Model checkpoint: French model, 150k training steps ("Boris") (9.0gb tar file!)
Cedille is a GPT-J model, it can be run using the mesh-transformer-jax
codebase.
The Cedille model is open-source:
- The model checkpoints are provided above (MIT License)
- It was trained using the
mesh-transformer-jax
codebase (Apache-2.0 License) - It can be run either using the 🤗Transformers library (Apache-2.0 License) or
mesh-transformer-jax
depending on your needs
And that's all there is to it! This repository is mostly here to guide you towards these resources 😉
Our paper is out now! https://arxiv.org/abs/2202.03371
Thanks for citing our work if you make use of Cedille
@misc{muller2022cedille,
title={Cedille: A large autoregressive French language model},
author={Martin M{\"{u}}ller and Florian Laurent},
year={2022},
eprint={2202.03371},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
Website: https://cedille.ai
Playground: https://app.cedille.ai/
Twitter: https://twitter.com/cedilleai
Paper: https://arxiv.org/abs/2202.03371
- The project was made possible with the generous support from the Google TRC program.
- Cedille is based on the GPT-J model developed by EleutherAI's wizards.