Hi, from here it looks like the inputs to a model/pipeline will need to be either the words or tokenizer output, which are dictionary outputs of id and mask. In my project, I need to use a combination of words + some extracted features. How would one do this? Thank you.
For example, I know that you can use “inputs_embds” with only continuous inputs. But is there a way to pass a combination of continuous input and discrete inputs?
No i think in that case you would implement a featurizer yourself and would pass the final
inputs_embeds to the model