In TensorFlow, a Session places the graph ops onto Devices, such as CPUs or GPUs, and provides methods to execute them. Is it possible to
It is possible to add new device types to TensorFlow, although we don't have great documentation. Approximately, it would require adding a new implementation of tensorflow::Device, adding a corresponding implementation of tensorflow::DeviceFactory, and invoking the following registration macro:
REGISTER_LOCAL_DEVICE_FACTORY("XPU", MyXPUDeviceFactory);