DRLM model for ranking

Hi,

I wanted to check if the DLRM model can be used for ranking instead of classification. I’ve updated the loss function accordingly, but the model doesn’t seem to be performing any better. I’m evaluating performance using NDCG@5, which is around 0. In contrast, XGBoost performs significantly better on the same data.

No files, tests, or entry points are named in the issue. Start by tracing DLRM's training and evaluation flow to verify how the updated ranking loss is connected, then reproduce NDCG@5 on the same data used by XGBoost. Done means determining whether ranking is supported and identifying the cause of the near-zero metric.

Written by the indexing model from the issue text.