[QST] Cross-entropy and pairwise losses are supported in Next Item Prediction

❓ Cross-entropy and pairwise losses are supported in Next Item Prediction

I found the text as following on page https://nvidia-merlin.github.io/Transformers4Rec/stable/model_definition.html?highlight=pairwise%20loss

Next Item Prediction: Predicts next items for a given sequence of interactions. During training, the prediction can be the next item or randomly selected items depending on the masking scheme. For inference, the intended purpose is to always predict the next interacted item. Cross-entropy and pairwise losses are supported.

Could you please tell me how to use pairwise losses instead of Cross-entorpy please?

Start with the Next Item Prediction section of the model_definition documentation page linked in the issue and trace how loss selection is exposed in the Python/PyTorch API. Document a concrete pairwise-loss usage example if the feature is supported, or clarify the limitation; done means the page answers how to use pairwise losses instead of cross-entropy.