It's not possible to specify the extra index URL when doing python -m pip install eland[pytorch], but we could instead ask to run the above command. In any case, it's an easy fix in the Docker image which is probably the main way that PyTorhc is used with Eland today (and #407 will make this even more true).
Installing all Eland dependencies locally or in the Docker image installs 4.2GB of dependencies, the two largest being PyTorch (1.9GB) and the CUDA dependencies (https://pypi.org/project/nvidia-cudnn-cu11/, https://pypi.org/project/nvidia-cublas-cu11/, https://pypi.org/project/nvidia-cuda-nvrtc-cu11/ and https://pypi.org/project/nvidia-cuda-runtime-cu11/, taking 1.4GB).
Since we don't support GPUs, installing those is wasteful, and we should install the PyTorch variant that only supports CPUs:
It's not possible to specify the extra index URL when doing
python -m pip install eland[pytorch], but we could instead ask to run the above command. In any case, it's an easy fix in the Docker image which is probably the main way that PyTorhc is used with Eland today (and #407 will make this even more true).