This model is trained as a part of InterIIT'21 competition, on the dataset provided by Bridgei2i. It is able to do multilingual (Hindi, English, Hinglish) summarization (many -> one) & is capable of generating summaries in English regardless of the input language.
| Rouge-L | Sacrebleu | Headline Similarity (using sentence-transformers) |
|---|---|---|
| p=0.46 r=0.49 f1=0.52 | 23.46 | 0.75 |
mBART is initialized from facebook/mbart-large-cc25 and is trained as per strategy mentioned in our GitHub.