Understanding "training" cell in tensorboard "graph"

Question

I created a keras model. I used tensorboard call back during model fitting and later used the tensorboard to visualize the model.

Attached are two images:

1. with model training enabled (default)
2. disable model training by setting "mode.trainable=False" before model compilation.

My question:

1. With disabling of model training some layers still have "trainable flag" along with the layer in the image (some of them are highlighted by red arrow).
2. Also there is a trainable cell in the top right of each image. what does it signify?

Ps: However i do see a difference in the number of trainable params is 0 when disabling training.

Total params: 35,511
Trainable params: 0
Non-trainable params: 35,511