This repository has been archived by the owner on Oct 9, 2023. It is now read-only.
Train general translation task #50
Labels
enhancement
New feature or request
help wanted
Extra attention is needed
won't fix
This will not be worked on
馃殌 Feature
Currently we use
mbart-large-en-ro
as defined here: https://github.com/PyTorchLightning/lightning-flash/blob/master/flash/text/seq2seq/translation/model.py#L39We should move towards
mbart-large
and pre-train from this backbone. This would be cleaner and more applicable to a large number of languages.The other consideration is that this model is around 600M parameters/2gb which is the reasonable size given the translation task. We may want to consider smaller variants for CI + quick iterations!
The text was updated successfully, but these errors were encountered: