Posted on 27th February 2024
As much as I know, it is the blend of relu and also softmax, which causes numerical problems, as relu can produce huge positive values matching to very tiny possibilities. If you modify your Model to work, say, tanh rather of relu for the latest dense layer, the difficulty will go away.
If you are looking for the Keras certification course, you can check out this online Keras Training and Improve your knowledge in Keras.
STILL GOT QUERIES?