More documentation needed on how to assess SSL training results

I am running the Self-Supervised Learning (SSL) task with SlowFast, and to monitor the training, on the Tensorboard, I am seeing the performance metrics:
-top5 error,
-top1 error and
-train/loss.

Since the SSL training uses no labels, how are these top-k errors calculated and how can I assess the trained model? Could you please provide more documentation for the SSL task?