Using tfjs-vis question

I am using tfjs vis to inspect a bit the model and the inner layers of the network.

There are 2 things I am trying to solve:

  1. What is the datatype of the weights here?
  2. How would I go about to determine whether this NN could potentially be downgraded to say float16 to reduce the size ?

Example (showing a summary of the net and the histogram of a layer)


There is no direct tfjs-vis API to get information about datatypes like core tensorflow in tfjs-vis but you can load tfjs model in and see model architecture with datatypes layer wise.

The default dtype is float32.

Thank you!

1 Like