mbart-50

This model is mbart-large-50-many-to-many-mmt model fine-tuned on the text part of SLUE-VoxPopuli spoken language understanding dataset.

The scores on the dev set are 83.25% and 87.76% for F1 and label-F1 respectively.