Finnish-NLP
/

byt5-base-finnish

@@ -97,16 +97,17 @@ Training code was from the Google's Jax/Flax based [t5x framework](https://githu
 Evaluation was done by fine-tuning the model on a downstream text classification task with two different labeled Finnish datasets: [Yle News](https://github.com/spyysalo/yle-corpus) and [Eduskunta](https://github.com/aajanki/eduskunta-vkk). Classification fine-tuning was done with a sequence length of 128 bytes.
-When fine-tuned on those datasets, this model (the fourth row of the table) achieves the following accuracy results compared to our other T5 models and their parameter counts:
 |                                                       | Model parameters | Yle News accuracy   | Eduskunta accuracy   |
 |-------------------------------------------------------|------------------|---------------------|----------------------|
 |Finnish-NLP/t5-tiny-nl6-finnish                        | 31 million       |92.80                |69.07                 |
 |Finnish-NLP/t5-mini-nl8-finnish                        | 72 million       |93.89                |71.43                 |
 |Finnish-NLP/t5-small-nl24-finnish                      | 260 million      |**94.68**            |74.90                 |
 |Finnish-NLP/byt5-base-finnish                          | 582 million      |92.33                |73.13                 |
 |Finnish-NLP/t5-base-nl36-finnish                       | 814 million      |94.40                |**75.97**             |
-|Finnish-NLP/t5-large-nl36-finnish                      | 1425 million     |TBA                  |TBA                   |
 Fine-tuning Google's multilingual mT5 models on the same datasets we can clearly see that our monolingual Finnish T5 models achieve much better results on Finnish text classification:

 Evaluation was done by fine-tuning the model on a downstream text classification task with two different labeled Finnish datasets: [Yle News](https://github.com/spyysalo/yle-corpus) and [Eduskunta](https://github.com/aajanki/eduskunta-vkk). Classification fine-tuning was done with a sequence length of 128 bytes.
+When fine-tuned on those datasets, this model (the fifth row of the table) achieves the following accuracy results compared to our other T5 models and their parameter counts:
 |                                                       | Model parameters | Yle News accuracy   | Eduskunta accuracy   |
 |-------------------------------------------------------|------------------|---------------------|----------------------|
 |Finnish-NLP/t5-tiny-nl6-finnish                        | 31 million       |92.80                |69.07                 |
 |Finnish-NLP/t5-mini-nl8-finnish                        | 72 million       |93.89                |71.43                 |
+|Finnish-NLP/t5-small-nl16-finnish                      | 184 million      |94.46                |74.00                 |
 |Finnish-NLP/t5-small-nl24-finnish                      | 260 million      |**94.68**            |74.90                 |
 |Finnish-NLP/byt5-base-finnish                          | 582 million      |92.33                |73.13                 |
 |Finnish-NLP/t5-base-nl36-finnish                       | 814 million      |94.40                |**75.97**             |
+|Finnish-NLP/t5-large-nl36-finnish                      | 1425 million     |94.17                |73.50                 |
 Fine-tuning Google's multilingual mT5 models on the same datasets we can clearly see that our monolingual Finnish T5 models achieve much better results on Finnish text classification: