Top-k Errors for SSL Training

I am running the SSL task with a _Contrastive Model_ and a _Contrastive Loss_ and as the performance metric, top-k errors are used. However, for an ssl task with a contrastive model, the labels are automatically dropped. Hence, how are top-k errors calculated for SSL tasks and are they still relevant metrics to monitor the performance?

If I run an SSL task on a data set with no labels, which metrics can I use to assess the training performance?