---
license: mit
language:
- ru
- en
library_name: transformers
tags:
- mbart
- mbart-50
pipeline_tag: text2text-generation
---

This is a smaller version of the [facebook/mbart-large-50](https://huggingface.co/facebook/mbart-large-50) with only Russian and English embeddings left.

sentencepiece vocabulary was shrinked from 250k to 25k (most common 10k English tokens and most common 15k Russian tokens). The creation of this model is heavily based on David Dale's method described [here](https://cointegrated.medium.com/how-to-adapt-a-multilingual-t5-model-for-a-single-language-b9f94f3d9c90), but with some specific to MBart additions.