Mohammed Innat

@MalyalaKarthik66 Thanks. I've just quicky checked the jax implementation, based on some condition, it also uses [for..loop](https://github.com/MalyalaKarthik66/keras/blob/4cc8ac0d17c4a5bb7658941816eaf9c20ff17aa0/keras/src/backend/jax/nn.py#L1504-L1513), no? Extending this for 3D cases later might cause issue, perhaps. Also, for...

@MalyalaKarthik66 I've just re-run my implementaiton with tf backend, works properly in tf backend. Looking at the error you face `'str' object has no attribute 'base_dtype'` seems like somewhere tf.ops...

@MalyalaKarthik66 If the issue originates from SKLearnClassifier or any component outside the PR, then addressing it would be beyond your scope. As long as the main implementation functions correctly, that...

@MalyalaKarthik66 It depends on the Keras team to move things forward. They’ll review to your PR as soon as they’re available. In the meantime, you can use the Gemini Code...

First of all, setting `built=1` is not needed and irrelevant here. About casting `ops.prod` to `int` - in that case I can just use `np.prod` instead of `ops.prod`. The behaviour...

@SamanehSaadat To reproduce the error.

@SamanehSaadat > The output of ops.prod doesn't work in layer's argument. Isn't this because its type is float? Are there other reasons? The issue is not with ops.prod itself. Thing...

@amitsrivastava78 Thanks for checking. They look similar but the way the function works underneath is different. Some discussion are made that mentioned ticket https://github.com/keras-team/keras-cv/issues/131#issuecomment-1133609670 - which lead to have this...

Might be related https://github.com/keras-team/keras/issues/19022 - pytorch-lightning like.

@pctablet505 Does the timing effect same between `model.test_on_batch` and `model.predict_on_batch`?