Conversation
|
I'll have a look at the LLPR arch, thanks |
I'd rather not do that, the user should give the unit of the data in the input to avoid this warning. Guessing |
|
This is just a dummy call that we do to |
|
cscs-ci run |
|
Thanks Paolo, your changes look good to me, so feel free to approve and merge or suggest changes if you want 👍 |
|
cscs-ci run |
|
cscs-ci run |
|
cscs-ci run |
|
Ok @ppegolo I have spent some time and I don't know how to hide the "fork" warning from pytest in a clean way so I added it back to the |
|
(in the CI it is fine, the logs don't contain the warnings) |
|
confirmed, no fork warnings |
Before the next release, we:
metatrain.utils.logging). These just warn about a problem that has been there since forever, it's not like the new release will be less robust than the previous ones in this regard.ModelOutput(quantity="X")to avoid getting metatomic's warning. These were useless anyway, so in principle it shouldn't be a problem. The only architecture that seems to useoutput.quantityis LLPR, so there I didn't change anything. But it could be that having removed it in PET will create problems. Could you see if things can be done without output.quantity @ppegolo ?unit="eV"to theModelOutputthat is initialized inmetatrain.utils.data.datasetto avoid getting an "empty unit" warning.📚 Documentation preview 📚: https://metatrain--1248.org.readthedocs.build/en/1248/