Albert Zeyer

The question is, those existing setups with Python 3.5/3.6, is there any good reason they use such old Python? Or could those setups just be switched to some newer Python?

> maybe start the process now I'm not really sure what you mean by that. Dropping Python will not be done in a branch or PR. This will be a...

> I mean to start the transition in i6_core _now_ and not only when the new OS is deployed. So you suggest the same here for RETURNN? Not wait until...

> I would like just to share one use case with python 3.8 and full-sum training, our HiWi Daniel had some additional scientific packages installed, and even though returnn run...

> no, as I said in the statement that you literally just quoted I suggest this for i6_core (well ok, also for other recipes). Ah ok, I was confused, because...

@Marvin84 Yes I'm very confident that this is not about the Python version, but rather about some mess-up in the system or environment. Esp as you point out that it...

> > some mess-up in the system or environment > > In which context do you think we should follow-up with more investigation on this? I think this is an...

I don't think GatherLayer is a good base. On Mon, Jun 14, 2021 at 11:00 AM Mikel Zhobro ***@***.***> wrote: > > https://github.com/rwth-i6/returnn/blob/6410df5eeb26cd9d77635a6e2af5a8330083198b/returnn/tf/layers/basic.py#L940 > > already offers the logic of...

I don't really understand why you would want that? I would assume this layer (like GatherNdLayer) only makes sense like this here. Do not care about other cases, esp not...

Can you post here what the actual problem/error is when you use these test cases with the original `slice_nd`? It is not really clear to me why there is actually...