---
language:
- uk
- en
tags:
- t5
---

The aim is to compress the mT5-large model to leave only the Ukrainian language and some basic English.

Reproduced the similar result (but with another language) from [this](https://towardsdatascience.com/how-to-adapt-a-multilingual-t5-model-for-a-single-language-b9f94f3d9c90) medium article. 

Results: 
- 1.2B params -> 779M params (37%)
- 250K tokens -> 8900 tokens
- 4.6GB size model -> 2.9GB size model