carant-ai/xlm-roberta-sentiment-base

Model Description

Finetuned xlm-roberta-base for Sentiment Analysis in English and Bahasa Indonesia

Training results

Trained on TPU VM v4-8 for ~3 hours

epoch	step	train_accuracy	train_loss	val_accuracy	val_loss
0	5391	0.955597997	0.118527733	0.963498533	0.098501749
1	10783	0.965486944	0.092906699	0.964689374	0.094814248
2	16175	0.968293846	0.085916176	0.965770006	0.093040377

Training procedure

For replication, go to GitHub page

Special Thanks

1. Google’s TPU Research Cloud (TRC) for providing Cloud TPU VM.
2. carlesoctav for making the training script on TPU VM
3. thonyyy for gathering the sentiment dataset