Keras 3 + Pruning?

[dbrandonk]

Does anyone know of any libraries that support Keras 3 + model pruning? tfmot does not support it and from what I can tell there is not a plan to tensorflow/model-optimization#1119 And all the other sparsity libraries that I have found only support Keras 2.

[Sameekshagithub]

Hi,

Currently, pruning support in the Keras 3 ecosystem is still limited compared to Keras 2.

tensorflow-model-optimization (tfmot) has historically been the main option for pruning, but Keras 3 compatibility depends on the backend and workflow being used.

Some possible approaches:

1. Use TensorFlow backend with Keras 3 and check the latest tfmot compatibility updates:
   https://github.com/tensorflow/model-optimization

2. If you only need sparsity for inference optimization, you can consider:

• structured pruning approaches implemented manually using Keras layers
• weight masking techniques
• exporting the model and applying optimization during deployment

3. If Keras 2 compatibility is acceptable, tfmot with TensorFlow/Keras 2 is still the most mature pruning solution.

It would also help to know:

• Which backend are you using (TensorFlow, JAX, or PyTorch)?
• Are you looking for unstructured pruning (weight sparsity) or structured pruning (removing filters/channels)?

That information may change the recommended approach.