[Question] Uniform query size for Listwise Ranking?

[kushagra-ch]

I was using the Listwise ranking section in the documentation as reference for a ranking task I'm working on.

In the documentation, the training data comprises of a list of 5 movies per training datapoint. For context, the data I'm working with has different sizes for each query, with some queries having 10 candidates to be ranked and others having >100 candidates to be ranked.

My question is, is it mandatory to fix the query size across all of the training datapoints, leading to each training sample in the dataset having a fixed query size or or can we train the model in such a way that each training sample has different query size, say by batching the training data based on the size of the query or using ragged tensors?

Thanks