Compute triplet loss as the differentiable activation to generate grad-weights?

Question

I\'m working on the problem to visualize embedding network using Grad-Cam in triplet-loss model as my base model. I\'m getting None values while computing,