Closed h-vetinari closed 3 years ago
Hi! This is the friendly automated conda-forge-linting service.
I just wanted to let you know that I linted all conda-recipes in your PR (recipe
) and found it was in an excellent condition.
Same problem as in #17: pytorch
from defaults has a problem with distributed
;
ImportError: cannot import name 'ReduceOp' from 'torch.distributed'
Torchvision now builds for OSX as well, meaning we can start looking at osx builds for master instead of 1.x
.
@conda-forge/pytorch-cpu @isuruf
This is now using the OSX builds of pytorch from conda-forge, but they yield the same error as the ones from the default channel:
from torch.distributed import ReduceOp
ImportError: cannot import name 'ReduceOp'
Have you ever seen this problem before?
@dirkgr
This now has all the dependencies but 56 tests fail. A lot of them seem to fail on module 'torch.distributed' has no attribute 'init_process_group'
- not sure if that is a pytorch API-difference between platforms, or if the conda-forge build forgot to build something.
Like I said on the other thread, I don't quite understand this. I run AllenNLP on Mac all the time, with the PyTorch version I get from Conda, and it works fine. Is it because I use conda install pytorch -c pytorch
?
Like I said on the other thread, I don't quite understand this. I run AllenNLP on Mac all the time, with the PyTorch version I get from Conda, and it works fine. Is it because I use
conda install pytorch -c pytorch
?
Hey @dirkgr, yes, this is the case because we cannot use other channels in conda-forge, and the conda-forge build apparently doesn't have distributed
. I don't know that package well enough, I'll open an issue on that feedstock.
I'll open an issue on that feedstock.
https://github.com/conda-forge/pytorch-cpu-feedstock/issues/46
With https://github.com/conda-forge/pytorch-cpu-feedstock/pull/47 merged, does this work now?
It seems I don't have the power to request re-running the tests.
With conda-forge/pytorch-cpu-feedstock#47 merged, does this work now?
Let's try.
It seems I don't have the power to request re-running the tests.
You do, but the azure page doesn't have a button for this functionality. You only see it when you're on the GH overview of the checks (which is where you arrive when you click on "Details"). Also, you should be able to close/open this PR, or simply write
@conda-forge-admin, please restart ci
At the time of #12 this was blocked on missing osx-support for
sentencepiece
, but that has materialized in the meantime.pytorch
will still have to be pulled fromdefaults
though, unfortunately.