warning

README.md

@@ -7,6 +7,8 @@ language: dv
 Pretrained from scratch on Dhivei (language of the Maldives)
 with ByT5, Google's new byte-level tokenizer strategy.
 
+**Use byt5-dv for now; this is less accurate**
+
 Corpus: Sofwath's Dhivehi corpus https://github.com/Sofwath/DhivehiDatasets
 
 Pretraining Notebook: 
@@ -17,3 +19,7 @@ https://colab.research.google.com/drive/1ERIZ1PyHn-yN_jo7dTQeODn22vrt-d1d?usp=sh
 On Dhivehi news classification task
 
 https://colab.research.google.com/drive/11u5SafR4bKICmArgDl6KQ9vqfYtDpyWp?usp=sharing
+
+## Issues
+
+There was an issue with the vocabulary size, final layer, and/or accuracy on fine-tuning.