File size: 647 Bytes
8684c70 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 |
---
language:
- sl
license: cc-by-sa-4.0
---
# t5-sl-large
t5-sl-large model is a Slovene T5 model. It has 24 encoder and 24 decoder layers, in total about 750 million parameters.
It was trained for 3 epochs on the following corpora:
## Corpora
The following corpora were used for training the model:
* Gigafida 2.0
* Kas 1.0
* Janes 1.0 (only Janes-news, Janes-forum, Janes-blog, Janes-wiki subcorpora)
* Slovenian parliamentary corpus siParl 2.0
* slWaC
## Evaluation
The model is described in detail and evaluated in our paper ["*Sequence to sequence pretraining for a less-resourced Slovenian language*"](https://arxiv.org/abs/2207.13988)
|