Matt Watson
Matt Watson
Thanks for filing! Will take a deeper look.
This is a known bug with some special classes we introduced for handling StaticHashTable and MutableHashTable in keras layers (TrackableWeightHandler and VocabWeightHandler). Both the classes have caused a number of...
I'm not sure a truly dynamic pairing of vocabulary indices and learned embedding weights (especially in a smoothly differentiable way) would be a good fit for either StringLookup or the...
@reedwm could you take a look here? Thanks!
@chenmoneygithub can you take a look here?
I don't think we would want to change the inputs to `on_train_end`, that would be disruptive to all users relying on the current behavior. If anything, the more general case...
@james77777778 will take a look soon! Sorry for the slow action this week, things are quiet with the US holiday.
Sorry for the delay on this! Valid bug! In fact we have some tests disabled on the torch backend for padding same for this reason it looks like. https://github.com/keras-team/keras/blob/61ea9c61f08f2a7b06091c63f032fc5e7839fb4f/keras/src/ops/nn_test.py#L1338-L1342 Marking...
@haifeng-jin any thoughts on this? I'm not familiar enough with the torch to know what our intended behavior is.
Just running you colab, I actually see a failure on all backends, jax/tf/torch. So this might just be a general bug. Either way, open for contributions if someone want to...