Sebastian Fischer
Sebastian Fischer
There is also: https://github.com/lorenzwalthert/touchstone
Are you working on a sequence to sequence modeling task, or what exactly are you trying to achieve?
Unfortunately, this is right now out of scope and I don't have the time right now to implement this, but we can keep this issue open as a feature request....
So you are interested in implementing this yourself?
I don't think there is much we can do about this on the `mlr3torch` side tbh. Different frameworks accessing cuda is always a pain. Are you running this on a...
The problem is that `repeat` is a keyword in R. We could either rename the function or document somewhere that it needs to be called like suggested by @rdinnage.
Ok, I realized that this is because `[.dataset` just calls into `$.getitem()` with whatever indices are provided. I am not sure what the correct behavior here is, but I think...
But the question is still whether `ds[1:2]` should throw an error if `ds` implements `$.getitem()`. The different tensors might have varying shapes, so it's not always possible to `torch_cat()` them.
Use the backend test helpers to verify that this is still a valid backend.
I also observe this