Conda environment : several environment files - specify cpu-only version of Pytorch

user_12237701 0
Viewed 1719

I am using conda 4.8.3 with Python 3.7, I am writing environment files to specify the dependencies of my project. I would like to write several files to be able to install several environments:

  • main.yml : containing the dependencies of my project, and pytorch CPU-only version
  • dev.yml : containing dev tools (mypy, flake8, pytest ..)
  • gpu.yml : containing pytorch-GPU (with a specified version of CUDA)

To get a basic (CPU) installation one would write : conda env update --file main.yml

To get an installation with GPU compatibility would then add conda env update --file gpu.yml

Here is my question : at the moment I cannot find the right way to specify the 'CPU-only' criteria for pytorch in an environment file, does anyone know if it is doable?

The command usually used for that purpose is conda install pytorch torchvision cpuonly -c pytorch, but I cannot find a way to specify it in the yml file.

On the pytorch channel site , there is a pytorch-cpu package, but its version is quite outdated (1.1.0, while the current main is 1.6.0)

Here is my main.yml environment file:

name: my_env
channels:
  - intel
  - conda-forge
  - pytorch

dependencies:

  - numpy
  - scipy
  - scikit-image
  - matplotlib
  - wxpython
  - colorama
  - dill
  - protobuf
  - pytorch   # How to specify the 'cpu' criteria here??
  - torchvision 
  - pip:
      - -r env/requirements.txt
pytorch conda python
1 Answers

In case anyone else is looking for the answer -- I tried what AMC suggested in a comment above. I can confirm that adding the line:

- cpuonly

to my environment.yml file forced the CPU version of pytorch to be downloaded.

user_13904890
Related