Hands-On GPU Computing with Python

Coupling Python with PyTorch for GPUs

There are basically two ways to get started with PyTorch for GPUs on your local system. The first one is to utilize a closed environment, such as Conda, that will keep your virtual environment concealed from the outer OS. The second way, with pip, is less preferable on a local system but can be useful for Google Colab (beginning the pip command with a ! prefix on the Colab Jupyter notebook).

Conda-based installation of PyTorch: As discussed in the previous section, this is the recommended way. To set up PyTorch within Anaconda, follow these steps.

First, let's install the pytorch, torchvision, and cudatoolkit=9.0 packages with Conda on a Terminal. Note that we used the same CUDA package for CuPy as ...