Bahasa T5 Model

Pretrained T5 base language model for Malay and Indonesian.

Pretraining Corpus

t5-base-bahasa-cased model was pretrained on multiple tasks. Below is list of tasks we trained on,

1. Unsupervised on local Wikipedia.
2. Unsupervised on local news.
3. Unsupervised on local parliament text.
4. Unsupervised on IIUM Confession.
5. Unsupervised on Wattpad.
6. Unsupervised on Academia PDF.
7. Next sentence prediction on local Wikipedia.
8. Next sentence prediction on local news.
9. Next sentence prediction on local parliament text.
10. Next sentence prediction on IIUM Confession.
11. Next sentence prediction on Wattpad.
12. Next sentence prediction on Academia PDF.
13. Bahasa SNLI.
14. Bahasa Question Quora.
15. Bahasa Natural Questions.
16. News title summarization.
17. Stemming to original wikipedia.
18. Synonym to original wikipedia.

Preprocessing steps can reproduce from here, Malaya/pretrained-model/preprocess.

Pretraining details

• This model was trained using Google T5's github repository on v3-8 TPU.
• All steps can reproduce from here, Malaya/pretrained-model/t5.

Load Pretrained Model

You can use this model by installing torch or tensorflow and Huggingface library transformers. And you can use it directly by initializing it like this:

from transformers import T5Tokenizer, T5Model

model = T5Model.from_pretrained('huseinzol05/t5-base-bahasa-cased')
tokenizer = T5Tokenizer.from_pretrained('huseinzol05/t5-base-bahasa-cased')

Example using T5ForConditionalGeneration

from transformers import T5Tokenizer, T5ForConditionalGeneration

tokenizer = T5Tokenizer.from_pretrained('huseinzol05/t5-base-bahasa-cased')
model = T5ForConditionalGeneration.from_pretrained('huseinzol05/t5-base-bahasa-cased')
input_ids = tokenizer.encode('soalan: siapakah perdana menteri malaysia?', return_tensors = 'pt')
outputs = model.generate(input_ids)
print(tokenizer.decode(outputs[0]))

Output is,

'Mahathir Mohamad'

Results

For further details on the model performance, simply checkout accuracy page from Malaya, https://malaya.readthedocs.io/en/latest/Accuracy.html, we compared with traditional models.

Acknowledgement

Thanks to Im Big, LigBlou, Mesolitica and KeyReply for sponsoring AWS, Google and GPU clouds to train T5 for Bahasa.