Nikita Shulga
Nikita Shulga
@pytorchbot merge
@pytorchbot merge -f "This code is not changed much"
@pytorchbot revert -m "Break cublas packaging into wheel" -c nosignal After the change `python -c "import torch;x=torch.rand(3, 3, device='cuda');print(torch.linalg.svd(torch.mm(x, x.t())))"` started to fail with ``` RuntimeError: Error in dlopen: /home/nshulga/miniconda3/envs/test-py37/lib/python3.7/site-packages/torch/lib/libtorch_cuda_linalg.so:...
Also, this change increased wheel size from 1.8G to 2.12G cc: @ZainRizvi , @seemethere we had some sort of mechanism for detecting binary size spikes, don't we?
/easycla
@mthrok deleting is easy, but we need to get to the bottom of what is publishing it
Deleted files and updated indicies... @atalman can you please check other domain libraries as well? (Also, quite surprised how it happened to audio, as I did all the M1 build...
Closing, as looks like it's been fixed now. Please do not hesitate to reopen/open new issue if it happens again.
Hmm, according to https://hud.pytorch.org/hud/pytorch/audio/nightly nightly builds are succeeding, but they are not getting uploaded to the channel for some reason Ok, I know the problem: we've provisioned much more machines...
Hmm, I see the same problem. Looks like something have change in the ecosystem: `conda install pytorch -c pytorch` attempts to install pytorch-1.10.2 rather than 1.11