I want to build a multilingual classification model (english, hindi, malayalam…) and wanted to ask if anyone has some suggentions which models to use. I would like to explore the performance of different models.
So which models are good in general for the usecase of a text classification with different languages?
These are all encoder-only Transformer models (great for classification, question-answering, NER,…).
CANINE is a relatively new model that is tokenizer-free, meaning it’s a character-level model and does not require an explicit tokenization step.
For summarization/translation/etc. (seq2seq tasks), mT5 is a great model.