François Chollet comments

Results 342 comments of


                                            François Chollet

Add parameter axis to tversky loss

Merged PR.

Support numpy.moveaxis operation

> This branch has conflicts that must be resolved Please resolve merge conflicts.

Using keras model with torch.func.jacrev gives zero gradients

> is there a workaround to make keras + Pytorch fully trasparent to pure pytorch code Keras is written in pure Python. When you are calling a Keras model with...

Fix stateful RNN to raise ValueError on batch size mismatch

You can put the test in `keras/src/layers/rnn/rnn_test.py` alongside `test_statefulness_two_states`. You can target `layer = layers.RNN(TwoStatesRNNCell(2), stateful=True)` as the RNN layer.

[Bounding Boxes] Add Support for Bounding Box Transformations during Image Resizing

Can you take a look at the test failures? I think they are likely related to the fact that torch tests are running with `data_format="channels_first"`. https://github.com/keras-team/keras/actions/runs/11469236802/job/31916066253?pr=20368

metrics.reset_state ignores shape of weights

This is expected behavior -- if you create any element of state, then you should also override `update_state` and `reset_state`. We could make the default behavior of `reset_state` more general...

Implementation of Ordinary Differential Equations

+1 to @rchao: we're open to contributions for such a feature. No plans on our side.

Add Flash Attention

For JAX, we may want to rely on Pallas. For TF, since we can't rely on custom ops, we may have to skip support. Presumably we should add it in...

Remote Model Loading/Saving + Docstring Fix

Fixed here: https://github.com/keras-team/keras/commit/b3e0b46323618cc1cf56715a9b9cf4baf2918a4b

Enormous memory usage after batched forward passes with TensorFlow 2.16.1 (CPU)

I tried running your snippet 20x and added a call to `gc.collect()` inside the loop. Here's what I get: ``` Memory usage after 0 run(s) (in MiB): 950.141 Memory usage...