Vocabulary and single image-question pair prediction

1. Is the vocabulary available that takes the words of the questions and converts them to 'input_ids'?
2. Is there a function that does this for an input question?
3. Is there a code that take a single image-question pair and predicts the answer?

Start by locating the repository's Python entry points for vocabulary handling and image-question inference, then check whether they expose the requested single-pair workflow. Done means identifying or adding a clear path for question-to-input_ids conversion and single image-question answer prediction, with usage documented for verification.

Written by the indexing model from the issue text.