Default behaviour in TextClassificationPipeline for regression problem type

[subhalingamd]

Feature request

The AutoModelForSequenceClassification class also supports problem_type=regression. I am not sure if it is popularly used but I believe that function_to_apply in TextClassificationPipeline should behave as if it is set to "none" for this case?

Motivation

Currently, the "sigmoid" function is also applied for regression tasks which, in my opinion, might not be the correct default behaviour.

Your contribution

I believe this can be handled by adding an additional condition for self.model.config.problem_type == "regression" before this line.