François Chollet

> Let me know what amount of operations I should cover in this PR to get it merged. So we will have the base OpenVINO support here and continue ops...

> @fchollet, will it be fine and sufficient for this PR merge? Yes, that sounds good to me!

When I say "integration tests", I was thinking that since the OpenVino backend is only supposed to support a specific set of use case, you could tests for these specific...

> for openvino backend, extend a list of ignore test files Sounds good -- so the changes would be limited to an OpenVino specific config file that lists the test...

The most explicit, idiomatic way would be to create an `Initializer` subclass. ```python @keras.saving.register_keras_serializable("my_package") class RollInitializer(keras.Initializer): def __call__(self, shape, dtype=None): return keras.ops.roll(keras.ops.arange(shape[-1]), shift=shape[-1] // 2) @keras.saving.register_keras_serializable("my_package") class SomeLayer(keras.Layer): def build(self,...

This is technically an invalid case (your output with TF has a size zero dimension, so it's an empty tensor). Weird that TF allows it -- but it's invalid nonetheless.

What's the performance like with JAX? Please try it.

The error "Functional has no attribute parameters" seems to suggest you're not using the torch backend. Try to print `keras.backend.backend()` and see what you get. Note that the original notebook...

Indeed, it looks like a deserialization bug. Mind you, it seems your code has some issues as well: 1. The layers added aren't built by the time you save the...

The Keras saving format isn't currently set up to save and load random seeds. That's not part of the API contract. It's something that could be added, though. For now...