Hi, there is a task that I am working on using a model with the transformer’s library. It is almost like a token classification task, but instead of assigning a label to each token, I want to predict a real number from 1-10 for each token. I wanted to ask what might be the easiest way to do this using the transformer’s library. I need to fine tune a pretrained model on my data.
It was tempting to think I might be able to use AutoModelForTokenClassification with 1 label, and so the model outputs a real number, which I thought I could scale appropriately. But then for training a TokenClassification task, the transformer’s library expects me to have labels for each token in the training data, not a number for each token in the training data.
Is there a quick and easy way to use something from the transformer’s library that can directly let me fine tune the pretrained model on my data for this kind of task?
Thank you!!