RuntimeError: PyTorch is not linked with support for opencl devices

RuntimeError: PyTorch is not linked with support for opencl devices When trying to set tensor.device(“opencl”)

There is no native support of OpenCL in pytorch - opencl device is just rudimentary device for something that was planned/existed at some time.

However you can use OpenCL for deep learning using out of tree backend.

There is a relatively new project based on OpenCL dlprimitives library that works on AMD, Intel and nVidia aritechures.

This project is relatively new and far from being complete but already gives support for training of many different vision network types and gives decent performance.

So follow the build instructions and try it.

See this discussion as well