---
language: es
license: CC-BY 4.0
tags:
- spanish
- roberta
pipeline_tag: fill-mask
widget:
- text: "Fui a la librería a comprar un <mask>."
---

This is a **RoBERTa-base** model trained from scratch in Spanish.

The training dataset is [mc4](https://huggingface.co./datasets/bertin-project/mc4-es-sampled ) subsampling documents to a total of about 50 million examples. Sampling is biased towards average perplexity values (defining perplexity boundaries based on quartiles), discarding more often documents with very large values (Q4, poor quality) of very small values (Q1, short, repetitive texts).

This model has been trained for 180.000 steps (early stopped from 250k intended steps).

Please see our main [card](https://huggingface.co./bertin-project/bertin-roberta-base-spanish) for more information.

This is part of the
[Flax/Jax Community Week](https://discuss.huggingface.co/t/open-to-the-community-community-week-using-jax-flax-for-nlp-cv/7104), organised by [HuggingFace](https://huggingface.co./) and TPU usage sponsored by Google.


## Team members

- Eduardo González ([edugp](https://huggingface.co./edugp))
- Javier de la Rosa ([versae](https://huggingface.co./versae))
- Manu Romero ([mrm8488](https://huggingface.co./))
- María Grandury ([mariagrandury](https://huggingface.co./))
- Pablo González de Prado ([Pablogps](https://huggingface.co./Pablogps))
- Paulo Villegas ([paulo](https://huggingface.co./paulo))