Determining probability of result when checking whether a premise and hypothesis are related
I am using the t5-large to determine whether a premise and hypothesis are related. That is whether the answer is an entailment, contradiction, or neutral.
I am trying to reproduce the first step of this paper (https://research.google/pubs/pub51307/) where they used the t5-11b and predicted whether a premise and hypothesis are related. They also calculated a score. For example, if a pair gave entailment, they also calculated P(entailment). They used the checkpoints from Google's repository directly.
I tried to find out whether this can be done using the model from HugginfFace. This is the code I am currently using. This just gives textual representation, whether something is entailment or contradiction, or neutral. I also was able to get some scores but not sure how to use them to calculate probability.
Any help to calculate the probability would be appreciated!
