Same problem. I am trying the following:
this link https://keras.io/getting_started/ suggests the compatbilities.
TensorFlow compatibility
The following Keras + TensorFlow versions are compatible with each other:
To use Keras 3:
tensorflow==2.15.0 & keras==3.0.0
tensorflow==2.16.0 & keras==3.0.0
Considering there is not tf-macos 2.16, I am using the first option with pip install.
There is this post I tried last week and it was working. I had to delete the venv, which was experimental, and now I can't replicate the result. I tried it because I also have the convergence problem, but I was able to solve it with Keras 3.0.0.
When I try to install Keras 3.0.0, I receive the following error:
"ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
tensorflow-macos 2.15.0 requires keras<2.16,>=2.15.0, but you have keras 3.0.0 which is incompatible."
I also tried to install Keras-nightly and at least it works. This are the result of a base ResNet50 with imagenet weights from tf.
Epoch 1/5
2/2 [==============================] - 14s 4s/step - loss: 6.4140 - accuracy: 0.3750 - val_loss: 9.1743 - val_accuracy: 0.5000
Epoch 2/5
2/2 [==============================] - 2s 355ms/step - loss: 1.2155 - accuracy: 0.7750 - val_loss: 9.4366 - val_accuracy: 0.7000
Epoch 3/5
2/2 [==============================] - 1s 601ms/step - loss: 4.1899 - accuracy: 0.6750 - val_loss: 539.9166 - val_accuracy: 0.6000
Epoch 4/5
2/2 [==============================] - 1s 571ms/step - loss: 0.3937 - accuracy: 0.8500 - val_loss: 1690.7887 - val_accuracy: 0.5000
Epoch 5/5
2/2 [==============================] - 1s 593ms/step - loss: 0.7857 - accuracy: 0.9750 - val_loss: 5639.2158 - val_accuracy: 0.5000
training time acceptable, training accuracy and loss okay. Still, there are problems with validation loss and accuracy, but it is a small step forward.
The pity is that I was able to make it work, but I had problems with Quarto for other reasons.
