Topic refinmenet
#3
by Harmpwns - opened
If I do for example a topic model and I get 5 documents and 10 keywords per topic, could I feed it to this model and get a nice label in Dutch? If so, where to start
Hi. This model needs finetuning on a task since this model has only been trained on trying to predict which tokens were masked out.
So you'd have to provide a dataset with examples showing input->output, and at least say 1000 of them.
Note that the t5 encoder/decoder models are kind of old, they haven't been updated with the latest advancements seen in decoder models, e.g. flash attention etc.