Conda channels

[pierreay]

Hi! I wanted to try the project. However, I'm not familiar with Conda. During the environment creation (using conda create) just as stated in the README, I got the following error:

  PackagesNotFoundError: The following packages are not available from current channels:

  - torchvision==0.15.0=py39_cu118
  - torchtriton==2.0.0=py39
  - torchaudio==2.0.0=py39_cu118
  - pytorch-mutex==1.0=cuda
  - pytorch-cuda==11.8=h7e8668a_3
  - pytorch==2.0.0=py3.9_cuda11.8_cudnn8.7.0_0
  - pyqtwebengine==5.12.1=pypi_0
  - pyqtchart==5.12=pypi_0
  - pyqt5-sip==4.19.18=pypi_0
  - pybind11==2.11.1=pypi_0
  - numba==0.56.4=np1.16py3.9h59fbc97_g288a38bbd_0
  - llvmlite==0.39.1=py39he1b5a44_0
  - libnvjpeg==11.9.0.86=0
  - libnpp==11.8.0.86=0
  - libcusparse==11.7.5.86=0
  - libcusolver==11.4.1.48=0
  - libcurand==10.3.2.56=0
  - libcufile==1.6.0.25=0
  - libcufft==10.9.0.58=0
  - libcublas==11.11.3.6=0
  - fastwer==0.1.3=pypi_0
  - cuda-runtime==11.8.0=0
  - cuda-nvtx==11.8.86=0
  - cuda-nvrtc==11.8.89=0
  - cuda-libraries==11.8.0=0
  - cuda-cupti==11.8.87=0
  - cuda-cudart==11.8.89=0

If I understand, you used additional Conda Channels that are not configured by default? If yes, I would be glad to see the proper Conda configuration (i.e., channels) that we have to use to successfully create the environment in the README. :)

[emidan19]

Hello @pierreay,

Were the other package requirements installed successfully? If so, while we check this bug, you can try using one of the following commands (with the conda environment already activated):
pip install <package>
or
conda install <package>
Where <package> would be pytorch, torchvision, torchaudio, pybind11, fastwer, etc. You may need to search for the proper way to install each library, specially using conda.

[pierreay]

Except the packages from the error above, the others installed correctly. I'm not familiar with Conda, but I expect it to install the packages automatically in the newly created environment, isn't it? By the way, I have a Miniconda distribution installed on my system. Since there is 200+ packages listed in requirements.txt, I'm not expecting to install all of them manually... :)

[troligtt]

I'm facing the same issue, any update?

[Jackal-Studios]

Overall fix:

0. Create and activate your conda environment without dependencies
1. Install the all the dependencies with pypi_0 from requirements.txt using pip install <package name>
2. Remove all the dependencies with pypi_0 from requirements.txt
3. Add the conda channels as explained below
4. Modify numba and llvmlite in requirements.txt
5. Run conda install --file requirements.txt

See below for further explanation:

0. Create and activate your conda environment without dependencies

conda create --name deeptempest
conda activate deeptempest

1. / 2. Deal with dependencies with pypi_0

For all the packages with a source specifier pypi_0 eg. pyqtwebengine==5.12.1=pypi_0 you will have to remove them from requirements.txt and run pip install <package name> Not ideal, but there are only ~5 of those.

• also quick note, I had to install pybind11 before installing other packages, so if you encounter any issues, try starting with installing pybind11.

3. Adding Conda channels by running the following commands:

conda config --add channels conda-forge
conda config --add channels pytorch
conda config --add channels nvidia

4. Modifying numba and llvmlite in requirements.txt

Lastly, looking at sources for numba and llvmlite, they did not contain releases with the specified hashes in requirements.txt. Therefore, you may want to try to modify the the requirements.txt

  - numba==0.56.4=np1.16py3.9h59fbc97_g288a38bbd_0
  - llvmlite==0.39.1=py39he1b5a44_0

  + numba==0.56.4
  + llvmlite==0.39.1

5. Run conda install --file requirements.txt !

Now all your dependencies should be installed.