Closed mattdangerw closed 2 weeks ago
Does this also get rid of the dependency on keras_core
?
@tirthasheshpatel yes, let me add that!
Going to rebase this and try it out (will just be nightly and github from HEAD for a bit). Please open an issue if this breaks you!
I'm curious about what the take_along_axis
error is.
Can we fix that in Keras?
This is just a draft to asses implications. At some point, KerasNLP will drop support for Keras 2, so we can focus on newer features in the library (multi-backend, model parallel training, quantization, etc).
Feel free to drop a comment if this would affect your workflow!