Matt Watson
Matt Watson
> Thought experiment: What if our users need to overwrite train_step, fit, etc for their own use case? Do they have to copy our boilerplate first? Could we provide a...
> We could also consider never preprocessing in model.call. This might feel more inconsistent to a user, but would cut down a lot of code here for sure. And maybe...
This was fixs on https://github.com/keras-team/keras-nlp/pull/1046
@ianstenbit what do you think?
@jonas-eschle agreed that adding this to the build rule sounds like the right call. Can you try that out? Thanks!
Talked this over, I am not sure we would want to go straight to a new layer at this time, but if you are interested, you could contribute a new...
@LukeWood could you take a look? Or get this to relevant CV person? To me it looks like this was a regression introduced with the newer KerasCV changes, but then...
Triage notes: we took a look and think that we should actually have concatenate throw an error with only a single input. We will try out a change and see...
I think we would want to try out an internal test run to see how many people this would break, so probably best handled by someone on the team directly?...
I think overall, `layers.pop` will not be supported on functional models. As generally speaking a functional model is a directed graph, a `pop` function is not really well defined. Sounds...