Once installed in your Python virtual environment, you can start working with Pytorch tensors in the DML virtual device. The key to using DirectML is to use a to(“dml”) command to run on your GPU.