Some Conda source download is too slow, use the command line conda install xxx
sometimes disconnected lead to library installation to fail. We can download by Thunder, the browser and other tools to specify dependent libraries, then install from the command line conda.
To pytorch1.1.0 example:
PyTorch 1.1.0 is a version released three months ago, and this version is no longer supported CUDA 8.0. Currently using the command line for installation, not only slow but also often interrupted, and the package is large (about 400M). Now installed by then downloaded to the local. Note the corresponding environmental python version cuda version.
pytorch1.1.0 Download: https://anaconda.org/pytorch/pytorch/files?version=1.1.0
Local installation package:
conda install --use-local $PATH_ROOT/XXX.tar.bz2