Does triplet loss seem very problematic?

Hi, I am using triplet loss for my training data. At the beginning of the training why it seems so bad? I am wondering whether this behaviour can be considered as a bad loss? Thanks in advance