audio
audio copied to clipboard
pytorch
Are the nightly conda builds broken?
13:35 $ conda install pytorch torchvision torchaudio -c pytorch-nightly
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
PackagesNotFoundError: The following packages are not available from current channels:
- torchaudio
Current channels:
- https://conda.anaconda.org/pytorch-nightly/osx-arm64
- https://conda.anaconda.org/pytorch-nightly/noarch
- https://repo.anaconda.com/pkgs/main/osx-arm64
- https://repo.anaconda.com/pkgs/main/noarch
- https://repo.anaconda.com/pkgs/r/osx-arm64
- https://repo.anaconda.com/pkgs/r/noarch
To search for alternate channels that may provide the conda package you're
looking for, navigate to
https://anaconda.org
and use the search bar at the top of the page.
Tested just now on OSX
Hi @greaber — are you still seeing this? I tried this just now on OS X and it worked fine.
Yeah, I just tried with a fresh conda env and python 3.9, and I get the same error
It seems that it is related to macOS version. The recent necessary migration on CircleCI silently bumped the minimum supported version to macOS 11. We need to fix this.
The minimum macOS version might be an issue for some people, but I don't think it is what I am seeing since I am on 12.4
@mthrok Do you mean if I install torch and torchvision using conda install pytorch torchvision -c pytorch-nightly?
In that case,
$ python -m torch.utils.collect_env
Collecting environment information...
PyTorch version: 1.13.0.dev20220719
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A
OS: macOS 12.4 (arm64)
GCC version: Could not collect
Clang version: 13.1.6 (clang-1316.0.21.2.3)
CMake version: Could not collect
Libc version: N/A
Python version: 3.9.12 (main, Jun 1 2022, 06:34:44) [Clang 12.0.0 ] (64-bit runtime)
Python platform: macOS-12.4-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
Versions of relevant libraries:
[pip3] numpy==1.22.3
[pip3] torch==1.13.0.dev20220719
[pip3] torchvision==0.14.0.dev20220719
[conda] numpy 1.22.3 py39h25ab29e_0
[conda] numpy-base 1.22.3 py39h974a1f5_0
[conda] pytorch 1.13.0.dev20220719 py3.9_0 pytorch-nightly
[conda] torchvision 0.14.0.dev20220719 py39_cpu pytorch-nightly
Yeah, nothing has changed. I haven't really tried debugging this myself since I don't know where to start, and it isn't that high a priority for me, but I'm happy to try any other suggestions someone may have
Looks like the M1 build job has been failing: https://github.com/pytorch/audio/actions/workflows/build-m1-binaries.yml
internal post: https://fburl.com/hknogcz7
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 now, but conda is not installed there, so it's not specific to just torchaudio, but likely affecting torchvision as well
@malfet I also get this issue with the main pytorch channel too. It was working fine a couple days ago. Using M1 macOS 12.4 / arm miniconda. Was there any change with the main pytorch channel ?
conda install pytorch torchvision torchaudio -c pytorch
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
PackagesNotFoundError: The following packages are not available from current channels:
- torchaudio
Current channels:
- https://conda.anaconda.org/pytorch/osx-arm64
- https://conda.anaconda.org/pytorch/noarch
- https://repo.anaconda.com/pkgs/main/osx-arm64
- https://repo.anaconda.com/pkgs/main/noarch
- https://repo.anaconda.com/pkgs/r/osx-arm64
- https://repo.anaconda.com/pkgs/r/noarch
To search for alternate channels that may provide the conda package you're
looking for, navigate to
https://anaconda.org
and use the search bar at the top of the page.
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
Could that be due to MacOS version ? Running on M1 with MacOS 11 seems to work for
conda install pytorch torchvision torchaudio -c pytorch
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: done
## Package Plan ##
environment location: /Users/ec2-user/miniconda3/envs/testenv
added / updated specs:
- pytorch
- torchaudio
- torchvision
The following packages will be downloaded:
package | build
---------------------------|-----------------
pillow-9.2.0 | py38h4d1bdd5_1 606 KB
torchvision-0.13.0 | py38_cpu 6.1 MB pytorch
------------------------------------------------------------
Total: 6.7 MB
The following NEW packages will be INSTALLED:
blas pkgs/main/osx-arm64::blas-1.0-openblas
brotlipy pkgs/main/osx-arm64::brotlipy-0.7.0-py38h1a28f6b_1002
bzip2 pkgs/main/osx-arm64::bzip2-1.0.8-h620ffc9_4
cffi pkgs/main/osx-arm64::cffi-1.15.0-py38h22df2f2_1
charset-normalizer pkgs/main/noarch::charset-normalizer-2.0.4-pyhd3eb1b0_0
cryptography pkgs/main/osx-arm64::cryptography-37.0.1-py38h834c97f_0
ffmpeg pkgs/main/osx-arm64::ffmpeg-4.2.2-h04105a8_0
freetype pkgs/main/osx-arm64::freetype-2.11.0-h1192e45_0
gettext pkgs/main/osx-arm64::gettext-0.21.0-h826f4ad_0
giflib pkgs/main/osx-arm64::giflib-5.2.1-h1a28f6b_0
gmp pkgs/main/osx-arm64::gmp-6.2.1-hc377ac9_3
gnutls pkgs/main/osx-arm64::gnutls-3.6.15-h887c41c_0
icu pkgs/main/osx-arm64::icu-68.1-hc377ac9_0
idna pkgs/main/noarch::idna-3.3-pyhd3eb1b0_0
jpeg pkgs/main/osx-arm64::jpeg-9e-h1a28f6b_0
lame pkgs/main/osx-arm64::lame-3.100-h1a28f6b_0
lcms2 pkgs/main/osx-arm64::lcms2-2.12-hba8e193_0
libgfortran pkgs/main/osx-arm64::libgfortran-5.0.0-11_2_0_he6877d6_26
libgfortran5 pkgs/main/osx-arm64::libgfortran5-11.2.0-he6877d6_26
libiconv pkgs/main/osx-arm64::libiconv-1.16-h1a28f6b_2
libidn2 pkgs/main/osx-arm64::libidn2-2.3.1-h1a28f6b_0
libopenblas pkgs/main/osx-arm64::libopenblas-0.3.20-hea475bc_0
libopus pkgs/main/osx-arm64::libopus-1.3-h1a28f6b_1
libpng pkgs/main/osx-arm64::libpng-1.6.37-hb8d0fd4_0
libtasn1 pkgs/main/osx-arm64::libtasn1-4.16.0-h1a28f6b_0
libtiff pkgs/main/osx-arm64::libtiff-4.2.0-h01837e1_1
libunistring pkgs/main/osx-arm64::libunistring-0.9.10-h1a28f6b_0
libvpx pkgs/main/osx-arm64::libvpx-1.10.0-hc377ac9_0
libwebp pkgs/main/osx-arm64::libwebp-1.2.2-h68602c7_0
libwebp-base pkgs/main/osx-arm64::libwebp-base-1.2.2-h1a28f6b_0
libxml2 pkgs/main/osx-arm64::libxml2-2.9.14-h8c5e841_0
llvm-openmp pkgs/main/osx-arm64::llvm-openmp-12.0.0-haf9daa7_1
lz4-c pkgs/main/osx-arm64::lz4-c-1.9.3-hc377ac9_0
nettle pkgs/main/osx-arm64::nettle-3.7.3-h84b5d62_1
numpy pkgs/main/osx-arm64::numpy-1.22.3-py38h25ab29e_0
numpy-base pkgs/main/osx-arm64::numpy-base-1.22.3-py38h974a1f5_0
openh264 pkgs/main/osx-arm64::openh264-1.8.0-h98b2900_0
pillow pkgs/main/osx-arm64::pillow-9.2.0-py38h4d1bdd5_1
pycparser pkgs/main/noarch::pycparser-2.21-pyhd3eb1b0_0
pyopenssl pkgs/main/noarch::pyopenssl-22.0.0-pyhd3eb1b0_0
pysocks pkgs/main/osx-arm64::pysocks-1.7.1-py38hca03da5_0
pytorch pytorch/osx-arm64::pytorch-1.12.0-py3.8_0
requests pkgs/main/osx-arm64::requests-2.28.1-py38hca03da5_0
torchaudio pytorch/osx-arm64::torchaudio-0.12.0-py38_cpu
torchvision pytorch/osx-arm64::torchvision-0.13.0-py38_cpu
typing_extensions pkgs/main/noarch::typing_extensions-4.1.1-pyh06a4308_0
urllib3 pkgs/main/osx-arm64::urllib3-1.26.9-py38hca03da5_0
x264 pkgs/main/osx-arm64::x264-1!152.20180806-h1a28f6b_0
zstd pkgs/main/osx-arm64::zstd-1.5.2-h8574219_0
I don't understand why it's blocking torchaudio installation on M1/macOS 12 while it was working perfectly fine a week ago, and no files have changed on the pytorch conda repo.
Here are the different combinations have tried: python 3.9 x86 on macOS 10.15: OK python 3.9 x86 on macOS 12.5: OK python 3.9 arm on macOS 12.4: FAIL python 3.9 arm on macOS 12.5: FAIL python 3.8 arm on macOS 12.5: FAIL
Unfortunately I have no way of testing python arm on macOS 11.
Hmm, I see the same problem. Looks like something have change in the ecosystem:
conda install pytorch -c pytorchattempts to install pytorch-1.10.2 rather than 1.11
Can repro that too.
Pushing investigations further, it seems it all comes down to the osx arm64 channel index being broken: https://conda.anaconda.org/pytorch/osx-arm64/current_repodata.json No torchaudio is listed and only 2 versions of pytorch are listed for arm64.
Still for some reason conda prefer to install pytorch 1.10.2 from the pkgs/main/osx-arm64 channel:
conda search pytorch -c pytorch
Loading channels: done
# Name Version Build Channel
pytorch 1.9.0.arm64 py3.8_0 pytorch
pytorch 1.9.0.arm64 py3.9_0 pytorch
pytorch 1.10.2 cpu_py310h36d6a73_0 pkgs/main
pytorch 1.10.2 cpu_py38h23cb94c_0 pkgs/main
pytorch 1.10.2 cpu_py39h23cb94c_0 pkgs/main
pytorch 1.11.0.arm64 py3.8_0 pytorch
pytorch 1.11.0.arm64 py3.9_0 pytorch
And the only reason conda install torchvision doesn't fail on arm64 is only because there's a very old (0.2.2) noarch version listed in the pytorch repo:
https://conda.anaconda.org/pytorch/noarch/current_repodata.json
Since torchaudio is not listed in osx-arm64 and no listed in noarch, it fails plain and simple.
Bottom line: https://conda.anaconda.org/pytorch/osx-arm64/current_repodata.json needs to be properly rebuilt so that pytorch and its ecosystem can work again on osx arm64/M1 computers.
Closing this as it was resolved in https://github.com/ContinuumIO/anaconda-issues/issues/13006#issuecomment-1202563268
@seemethere
I just tested now but the issue is not resolved. I am on M1 mac.
conda install pytorch torchaudio -c pytorch-nightly
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
PackagesNotFoundError: The following packages are not available from current channels:
- torchaudio
Current channels:
- https://conda.anaconda.org/pytorch-nightly/osx-arm64
- https://conda.anaconda.org/pytorch-nightly/noarch
- https://conda.anaconda.org/conda-forge/osx-arm64
- https://conda.anaconda.org/conda-forge/noarch
To search for alternate channels that may provide the conda package you're
looking for, navigate to
https://anaconda.org
and use the search bar at the top of the page.
The stable release is working, but the original issue of nightly is not resolved.
conda install pytorch torchaudio -c pytorch
Collecting package metadata (current_repodata.json): done
Solving environment: done
## Package Plan ##
environment location: /Users/moto/miniforge3
added / updated specs:
- pytorch
- torchaudio
The following NEW packages will be INSTALLED:
torchaudio pytorch/osx-arm64::torchaudio-0.12.0-py39_cpu
The following packages will be SUPERSEDED by a higher-priority channel:
pytorch pytorch-nightly::pytorch-1.13.0.dev20~ --> pytorch::pytorch-1.12.0-py3.9_0
Proceed ([y]/n)?
If it helps with the investigation of this issue - the pytorch channel is mirrored and served from our CDN, but the pytorch-nightly channel is not. So, for this I think we would need to look into what's going on with anaconda.org in the pytorch-nightly repodata.
This looks like resolved. Yesterday, there was no torchaudio for osx-arm64 in nightly channel, but today there is. Perhaps by https://github.com/pytorch/audio/pull/2621.