I want a cpu version of Pytorch > 0.2 with Python 2.7. Given all these constraints do we plan to have different versions?
Also, was wondering why is the size of the Pytorch image (docker pull pytorch/pytorch:latest) so hug ~3.5G.
Also, I am trying to write my own Dockerfile. However, the following commands prompts the user to proceed or not (with a list of packages it installs) and therefore requires manual intervention:
The Docker PyTorch image actually includes everything from PyTorch dependencies (numpy pyyaml scipy ipython mkl) to the PyTorch package itself, which could be pretty large because we built the image against all CUDA architectures. The CPU version should take less space.
We can use conda install pytorch-cpu torchvision -c pytorch to install PyTorch > 0.2 with Python 2.7. To automate this process, we can use the -y option, so the command would look like: conda install -y pytorch-cpu torchvision -c pytorch