François Chollet
François Chollet
> The Keras saving format isn't currently set up to save and load random seeds. To provide a bit more info here -- saving and loading work by walking the...
In this case, to check forward pass correctness, you can simply call the model in inference mode (`training=False`).
Thanks for the PR! Have you tried to time it on GPU compared to regular attention? I was under the impression that we were going to need a custom Pallas...
An alternative contribution has been merged. Thanks for the PR in any case!
The first step (or two steps sometimes) cover compilation time too (which is why they tend to take much longer). The other steps cover a full forward pass, which includes...
> By first step you mean calling the model.predict for the first time (or second, as you said) in the given session? `predict()` (and fit, evaluate) are loops. The inner...
> How to fix this test case any idea ? add(mx_array, numpy_array) works but fails when using + operator. Should we skip this test for mlx backend ? It's not...
Please note, the nn and rnn namespaces are the most important for getting mlx to work with typical workflows. On Fri, Jan 17, 2025, 4:23 PM acsweet ***@***.***> wrote: >...
Right, unless mlx actually exposes some cudnn bindings for these On Fri, Jan 17, 2025, 11:15 PM acsweet ***@***.***> wrote: > I'm going to hold off on math.qr for now,...
Your custom objects always need to be serializable. You can use `@keras.saving.register_keras_serializable` for this purpose https://keras.io/api/models/model_saving_apis/serialization_utils/#registerkerasserializable-function