François Chollet
François Chollet
I don't understand the connection. You could try pruning things from your change until the test passes, then you'll have a good idea what particular lines are causing the issue.
> specific in torch backend, when a list[Layer] is observed, wrap it in nn.ModuleList() and then wrap it with TorchModuleWrapper(). this might work but i will need to double check...
I would recommend not adding any new API. What about this approach: > (doesn't seem to work) specific in torch backend, when a list[Layer] is observed, also call self.register_module() in...
> E AssertionError: Items in the first set but not the second: What are those `_layer_n` attributes? > i think the main issue to me is that deserialization don't triggers...
Thanks for the update! There seems to be one failing test: https://github.com/keras-team/keras/actions/runs/10128018370/job/28006468053?pr=19908
Thanks for the PR! > I opted to try to maximize support for N-D searchsorted (because this is my use-case). However, numpy does not support it. JAX supports it by...
> In that case, should we raise an error if the user passes an N-D sorted_sequence, or let the backend handle it if it is incompatible? Better to do it...
Rather than using the function `ndim()` you could use `len(shape(x))`, which works the same everywhere?
When you are using `steps_per_epoch` or `validation_steps`, you keep drawing from the same generator across different epochs. At epoch 2, you have no data since you already consumed all the...
In the example above, `z_in` has a `dtype` attribute which in this case has value `"float32"`.